Managerial Duties and Managerial Biases∗

Ulrike Malmendier†, Vincenzo Pezone, Hui Zheng

UC Berkeley, Goethe University, and UC Berkeley

Abstract

Much of the evidence on managerial biases in corporations focuses on the role of the CEO. Weargue that analyzing individual managers in isolation can result in misattribution. In the contextof financing decisions, we show that overoptimistic beliefs of the CFO, rather than the CEO,are a significant determinant of pecking-order distortions and high leverage. CEO overconfidencematters indirectly as the CEO’s and not CFO’s type determines investors’ assessment of default risk.Moreover, overconfident CEOs tend to hire overconfident CFOs whenever given the opportunity,generating a multiplier effect.

∗We would like to thank colleagues and seminar participants at the University of California Berkeley, University of Bonn,University of Chicago, CSEF (University of Naples), Northwestern University, Stanford University, and Yale University as well asthe American Finance Association and Adam Smith Conference in Oxford for helpful comments. Canyao Liu, Jana Willrodt, andJeff Zeidel provided excellent research assistance.†Corresponding author. Department of Economics and Haas School of Business, 501 Evans Hall, Berkeley CA 94720; phone:

510-642-5038, fax: 510-642-6615; email: ulrike@berkeley.edu.

1 Introduction

Managerial biases, and especially managerial overconfidence, appear to have significant explanatory power for

corporate decisions. Earlier work such as Hambrick and Mason (1984) have related organizational outcomes

to personal traits of top managers. Recent empirical work has established a significant role of managerial

biases in shaping investment, merger, and financing decisions (see, e.g., the overview in Baker and Wurgler

(2013)). The spectrum of managerial traits and biases considered in the corporate-finance literature ranges

from risk aversion, education, childhood experiences, and gender, to behavioral biases such as overconfidence,

loss aversion, and escalation of commitment.1 Kaplan, Klebanov, and Sorensen (2012) argue that these traits

and biases have a first-order impact on corporate performance. Behavioral corporate finance, and in particular

research on managerial biases, is currently the fastest-growing strand of behavioral-finance research.2

Much of this research focuses on the traits and biases of one type of manager, typically the chief executive

officer (CEO). The emphasis on CEOs reflects both their central role as the top decision makers and, more

mundanely, data availability. Few papers touch on the roles of other top managers, such as the chief financial

officer (CFO), and even less research considers two or more top mangers jointly.3 In this paper, we argue that it

is important to account for managers other than the CEO when assessing the magnitude and empirical relevance

of managerial biases. Otherwise, both theoretical and empirical analyses run the risk of misattribution. Such

misattribution is particularly likely under assortative matching of managers who share similar belifs.

We illustrate this point in the context of financing decisions. Specifically, we jointly assess the influence of

CEO and CFO overconfidence on the choice of external financing and on the financing conditions offered by

investors. The context of financial decisions provides us with a setting where two types of managers plausibly

exert a large influence and are easily identified as the most relevant decision-makers: the CEO since she has

1 See Graham, Harvey, and Puri (2013), Bertrand and Schoar (2003), Malmendier and Tate (2005 and 2011), Malmendier et al.(2011), Chevalier and Ellison (1997), Huang and Kisgen (2013), Faccio, Marchica, and Mura (2016), Yim (2013), Camerer andMalmendier (2012), Bazerman and Neale (1992), and Ross and Staw (1993), among others.

2 Malmendier (2018) shows the publication growth rates of different types of behavioral-finance research in top finance andeconomics journals, including the explosion of research on managerial biases and social ties over the last years (31-70% growth,compared to 9-12% in other fields of behavioral finance research).

3 Notable examples of CFO studies include Ben-David, Graham, and Harvey (2007), Ben-David and Graham (2013), Jiang,Petroni, and Wang (2010), and Chava and Purnanandam (2010). Studies that analyze several of the C-suite managers includeAggarwal and Samwick (1999), Datta, Iskandar-Datta, and Raman (2001), and Selody (2010).

1

the ultimate say and determines the riskiness of the firm’s projects, and the CFO since financial activities and

operations are his core responsibility (see, e.g., Berk and DeMarzo (2007)).4 And the focus on overconfident

beliefs as the managerial bias reflects that overconfidence is the most extensively researched and most robustly

documented behavioral trait in corporate decision-making, or “the mother of all biases,” as Bazerman and

Moore (2013) put it.5

We define managerial overconfidence as managers’ overoptimistic belief about the future returns, or cash

flows, accruing to their firms. As our empirical proxy, we employ the widely used Longholder measure. It

captures managers’ personal overinvestment in their firm in the form of delayed option exercise (see, e.g., the

overview in Malmendier and Tate (2015)). The basic idea is simple: Managers who overestimate future returns

to their firm believe that there will be significant increases in firm value in the future. One way to benefit from

future stock-price increases is not to exercise in-the-money executive stock options. That is, overconfidence

leads managers to delay exercise. While rational managers typically exercise executive stock options some

years before expiration, depending on how much the options are in the money, overconfident managers tend to

delay exercise in order to reap the future increases in value. As has been shown before, and as we will replicate

here, the proxy can be distinguished from insider trading, signaling and other alternative interpretations.

We are the first to apply this measure to top managers beyond the CEO. Note that the interpretation

of late option exercise varies to some extent across CEO and CFO. In the case of the CEO, a Longholder

has overoptimistic belief about her abilities to generate returns. In the case of the CFO, a Longholder also

overestimates future returns. However, his overconfident beliefs likely stem from (also) overestimating the

potential of the firm and the CEO’s ability to generate returns. In other words, the CFO Longholder proxy

reflects overoptimistic beliefs in the firm and in another person (the CEO), rather than only the CFO’s own

ability to generate returns. Despite these potential differences, we stick to a common label, overconfidence, to

describe the bias associated with a common empirical measure, late exercise of executive stock options.6

4 Our approach can be applied to other C-suite managers and their responsibilities, e.g., the COO and operating decisions.The intersection of ExecuComp and Thompson data is currently too small to perform such an analysis in our data; see Section 3.

5 More than half (53%) of all papers on managerial biases in top finance and economics journals analyze overconfidence (Mal-mendier (2018)). See also Meikle, Tenney, and Moore (2016) for a survey of the large research on the organizational consequencesof overconfidence in firms.

6 For our empirical analyses it is not of first-order importance precisely why the CFO is overconfident since the Longholdermeasure captures return overestimation regardless of the reason. In fact, the empirical proxy could also be used for a theoretical

2

We start from a simple model to generate empirical predictions on financial decisions. The model differs

from previous work on CEO overconfidence (Malmendier and Tate (2005; 2008)) on two dimensions. First, we

consider the possibility of both the CEO and the CFO exhibiting overconfidence. Second, we allow the CEO’s

optimistic beliefs to affect her effort. The model illustrates that overconfidence can induce a CEO to exert

more effort than a rational CEO, and how the CFO, in turn, accounts for such behavior in his financing choice.

The model generates three main testable predictions. First, holding constant the CEO’s type, an over-

confident CFO exhibits a preference for debt over equity (pecking order) when accessing external finance.

Intuitively, overconfident CFOs believe that the broader market underestimates the stream of future cash flows

generated by the CEO’s investment choices and thus the value of their firm. Since equity prices are more

sensitive to differences in opinions about future cash flows, overconfident CFOs find equity financing too costly

and “even more overpriced” than debt. This argument is similar to the prediction for CEOs in Malmendier

et al. (2011), with the difference that we allow the CFO to choose the means of financing.7

Second, an overconfident CEO exerts a significant indirect influence on financing as overconfidence can

lower the cost of financing. The reason is that the CEO’s optimistic beliefs about the returns to effort induce

higher effort, similarly to the mechanisms in Pikulina, Renneboog, and Tobler (2014) or Gervais, Heaton, and

Odean (2011). Following a negative shock, a rational CEO is less willing to work hard than an overconfident

CEO, who is optimistic enough to work towards the good outcome regardless. In this case, overconfidence

helps solve the incentive problem. Anticipating such behavior, debtholders require a lower premium on debt

from an overconfident than from a rational CEO.8

This second prediction allows for a further refinement, in terms of empirical testability, in that it pinpoints

the type of firm that is predicted to drive the association between CEO overconfidence and the cost of debt.

Namely, the association should vary non-monotonically with the firm’s profit variability: Very mild shocks to

profits do not matter much for the effort choices of either type of CEO, and severe shocks diminish the incentives

concept of overconfidence that incorporate overoptimism about own ability on the side of the CFO. The different plausibletheoretical approaches do not affect our results.

7 Empirically, we also analyze the role of the CEO in determining the type of financing, and show that it is the belief of theCFO, not the CEO, that matters.

8 Designing different contracts based on the CEO’s degree of bias requires stakeholders be able to recognize managerialoverconfidence. Otto (2014) argues that this assumption is supported by the empirical evidence.

3

to work for both types of CEOs. For some intermediate range, however, a rational CEO anticipates the project

to be out of the money and does not exert effort, while an overconfident CEO overestimates the return to

effort enough to work hard. This non-monotonicity is specific to the model of biased beliefs, and helps rule

out alternative explanations based on omitted firm characteristic. It also implies that the favorable financing

conditions offered to overconfident managers do not merely reflect investors assessing the CEOs’ abilities

too positively, as shown experimentally by Schwardmann and van der Weele (2017). While it is a plausible

additional channel for our second findings that confident managers mislead investors, non-monotonicity implies

a rational response of investors when offering the favorable financing conditions: Investors correctly anticipate

that an overconfident CEO will put all her effort behind the project she is engaging in.

A third prediction is that, as another indirect channel, CEO overconfidence affects financing through hiring.

We show that, when selecting a new CFO, an overconfident CEO is more likely to choose one who shares her

views regarding the firm’s profitability. While CEOs do not have complete discretion in hiring a new CFOs,

they do have a significant say in the selection of board members, who in turn choose the CFO (Shivdasani

and Yermack (1999); Cai, Garner, and Walkling (2009); Fischer, Gramlich, Miller, and White (2009)). This

prediction implies a potential multiplier effect of overconfident managers.

All predictions find strong support in the data. We first replicate the Longholder measure from Malmendier

et al. (2011)9 and generate a parallel CFO measure. The Longholder proxy measures delayed option exercise

relative to a benchmark model of optimal exercise of executive stock options (Hall and Murphy (2002)). As a

robustness check, we also construct a continuous version of our Longholder proxy following Otto (2014).

We then test the first prediction, on the choice between debt and equity. We use various measures of net

debt issuance from Compustat, SDC, as well as traditional financing-deficit models. We find that overconfident

executives are reluctant to issue equity. Relatedly, we estimate a positive association between overconfidence

and leverage choices. In both sets of estimations, CFO overconfidence is statistically and quantitatively more

important than CEO overconfidence and, if analyzed jointly, CEO overconfidence is insignificant in our data.10

In other words, optimistic beliefs of both the CEO and the CFO predict that external financing is tilted

9This measure corresponds to the “Longholder Thomson” variable in Malmendier et al. (2011).10 Note that our estimates are not directly comparable to the earlier literature in that our more recent data does not reveal a

strong CEO effects in capital structure decisions to begin with, even when neglecting the CFO.

4

towards debt, but the CFO’s beliefs strictly outweigh those of the CEO. Thus, the manager whose beliefs

matter for capital budgeting decisions directly appears to be the CFO, which is consistent with the CFO’s core

compentency in designing the financing model and with prior evidence on the CFO being in charge of financing

decisions. For example, Bertrand and Schoar (2003) evaluate the relative importance of CEO and CFO fixed

effects and conclude that CFOs appear to dominate CEOs in financial policies. Graham, Harvey, and Puri

(2015) provide survey evidence that CEOs are more likely to delegate capital structure decisions than, for

examples, those related to mergers and acquisitions. Vice versa, a significant fraction of CFOs in their survey

claim to make capital structure with little or no input from other executives. Turning to the role of beliefs,

Ben-David and Graham (2013) find that CFOs’ beliefs about the volatility of future stock-market returns have

a significant explanatory power for leverage decisions.11 We bring this characterization of the role of the CFO

to the realm of behavioral biases, and show that his overconfident beliefs dominate the beliefs of the CEO.

To test the second model prediction, on the cost of financing, we merge DealScan data on syndicated loans

with our data set. We find that firms with overconfident CEOs obtain significantly better financing conditions,

as measured by the interest rates on their corporate loans. Overconfident CEOs are charged significantly

lower rates, controlling for the known determinants of the cost of debt. This is consistent with the CEO

being responsible for project implementation and completion, and investors inferring her willingness to push a

project through even in the presence of negative shocks. Hence, CEO overconfidence exhibits a strong indirect

influence on financing by affecting investors’ assessment of risk and the resulting cost of financing.

Moreover, the effect is non-monotonic, following the pattern predicted by our model: It is significant only

for firms characterized by medium earnings variability. This result holds regardless of whether we use earnings

volatility, analyst coverage, or analyst forecast variability as proxies for profit variability, and it is robust

over a broad range of plausible cutoff points for the ‘intermediate’ range. The pattern of non-monotonicity

is consistent with the explanation offered by our model, namely, differences in effort provision in bad states

of the world. Investors anticipate that overconfident CEOs will push ahead when rational managers start to

11The view that CFOs have a major influence on financing decisions is widely accepted also beyond academia. For ex-ample, according to the popular educational website for practitioners investopedia.com, “the CFO oversees the capital struc-ture of the company, determining the best mix of debt, equity and internal financing. Addressing the issues surroundingcapital structure is one of the most important duties of a CFO.” (Source: https://www.investopedia.com/ask/answers/

what-does-chief-financial-officer-cfo-do/)

5

divert effort. The result also addresses concerns about alternative, unobserved determinants of the cost of

debt. Any such alternative determinant would need to exhibit the same non-monotonicity, and would need to

be correlated only with the CEO Longholder variable, not the CFO Longholder variable.

Third, we test who is likely to be chosen as CFO conditional on the beliefs of the CEO. Based on our

theoretical prediction, we expect the commonality of beliefs to play an important role. Indeed, we find that

companies with overconfident CEOs are more likely to appoint like-minded CFOs. The statistical and economic

magnitudes of the effect are large, confirming the presence of assortative matching. Thus, CEOs exert indirect

influence on corporate financing also via their influence on CFO selection.

Overall, our findings confirm the thrust of the existing literature in that they provide evidence on the

significant role of managerial biases for corporate outcomes. By focusing on the CFO and showing that his

beliefs affect outcomes in his domain, we help complete the literature on managerial overconfidence, which had

focused on the CEO. The domain specificity of the bias, i.e., the relevance of CFO overconfidence for financing

and of CEO overconfidence for other managerial tasks, corroborates the empirical importance of managerial

overconfidence and implicitly the interpretation of the widely used Longholder proxy.

Our results caution against the focus on one single manager in the analysis of corporate decisions, including

behavioral analyses. Our results suggest that previously identified effects of CEO overconfidence could reflect

biases of the CFO – though with the caveat that our data does not suggest strong CEO effects on capital

structure to begin with, even when omitting the CFO. This may reflect a higher degree of delegation of

financing decisions from CEOs to CFOs in more recent years. Empirical analyses of managerial traits and

biases would benefit from more complete firm data sets that include all top managers who influence a specific

firm outcome. As a practical implication, the joint consideration of managerial biases is important for boards

when composing the C-suite and when devising corporate-governance responses to biased managerial behavior.

Literature Review. Our analysis relates to previous work on the role of CFOs and their biases, including,

Ben-David et al. (2007), Ben-David and Graham (2013), Jiang et al. (2010), and Chava and Purnanandam

(2010). Using the methodology of Bertrand and Schoar (2003), Ge, Matsumoto, and Zhang (2011) find that

CFO “style” is related to accounting choices. Huang and Kisgen (2013) link the gender of CEOs and CFOs

6

to the returns to acquisitions (where male executives are likely to be more overconfident). More broadly, our

paper extends the literature that shows that differences in beliefs between investors and managers affect capital

struture choices (Dittmar and Thakor (2007) and Boot and Thakor (2011)). Outside the behavioral realm,

Jiang et al. (2010) and Kim, Li, and Zhang (2011) show that CFOs’ equity incentives have stronger explanatory

power for earnings management and stock crashes than those of CEOs. We are the first to bring the comparison

of CFO and CEO traits for financing decisions to the realm of overconfidence, to jointly consider overconfidence

among different managers, and to analyze direct and indirect channels of influence.

Our paper also extends the literature that links overconfidence to capital-structure decisions. The survey ev-

idence in Graham and Harvey (2001) suggests that CFOs’ reluctance to issue equity may reflect overconfidence.

From a theoretical perspective, Hackbarth (2009) predicts higher debt ratios for managers who overestimate

earnings growth. Landier and Thesmar (2009) and Graham et al. (2013) confirm empirically that overconfi-

dence is associated with higher leverage and a preference for short-term debt. Consistent with prior work, our

model also connects overconfidence with higher debt ratios; but we show that overconfidence at the CFO level,

rather than at the CEO level, matters most in this context.

Our second set of results, on the terms of financing, contributes to the literature emphasizing the “bright

side” of overconfidence. Ever since the influential paper by Roll (1986) on the link between managerial “hubris”

and poor returns to acquirers, it has been a puzzle why boards continue to appoint overconfident managers,

who make poor decisions in a host of contexts (see the overview in Malmendier and Tate (2015)). More recent

papers argue that overconfident managers may increase firm value (Goel and Thakor (2008)), innovate more

(Hirshleifer, Low, and Teoh (2012)), and require lower levels of incentive compensation for a given amount of ef-

fort (Otto (2014)). Others argue that mild overconfidence can prevent underinvestment (Campbell, Gallmeyer,

Johnson, Rutherford, and Stanley (2011)), reduce conflicts between bondholders and shareholders such as the

debt overhang problem (Hackbarth (2009)), or can be advantageous in oligopolistic market (Englmaier (2010,

2011)). We provide a new angle on the “bright side” of overconfidence by showing that overconfident CEOs

obtain lower interest rates on corporate loans. The proposed mechanism is that overconfident CEOs exert more

effort, in line with Gervais and Goldstein (2007) and Hilary, Hsu, Segal, and Wang (2016). We also identify the

firms that benefit most from hiring an overconfident manager – those experiencing medium profit variability.

7

Our results on better financing conditions for overconfident CEOs are also consistent with the experimental

evidence that people rate an overconfident person as significantly more likely to be successful (Schwardmann

and van der Weele (2017)). However, the non-monotonicity in return variability points to the effort angle,

rather than the mere ability of overconfident CEOs to persuade others of their success chances.

Finally, our model relates to studies of dissent between managers in organizations (Landier and Thesmar

(2009); Landier, Sauvagnat, Sraer, and Thesmar (2013)), which suggest that CEOs are more likely to hire

like-minded executives. Relatedly, Graham et al. (2015), Acemoglu, Aghion, Lelarge, Reenen, and Zilibotti

(2007), and Bloom, Sadun, and Van Reenen (2012) analyze empirically when and where managers are likely

to delegate. In the Graham et al. (2015) survey, “gut feel” is an important element for 48.2% of the CEOs

in the decision to delegate corporate investment tasks to lower-level executives. Our empirical results support

the inclination to lean on “like-minded” people in the context of managerial overconfidence. They are also

consistent with the finding in Goel and Thakor (2008) that overconfident managers are more likely to be

appointed as CEOs. Here, we ask who is chosen as CFO conditional on the overconfidence of the CEO, and

show that the commonality of personal traits plays an important role.

2 Theoretical Framework

We consider a simple model of investment and financing, following closely the standard textbook treatment

of Tirole (2006). Our key modification relative to previous work on managerial overconfidence is to assign

decisions related to capital structure to the CFO. This assumption naturally delivers a relation between CFO

overconfidence and a preference for debt, and can accommodate a link between CEO bias and hiring decisions

and lending terms in a straightforward way. We discuss how our conclusions generalize to a less simplistic

set-up. Our main emphasis lies on providing a consistent unifying underpinning for our empirical tests.

2.1 Setting of the Model

The role of the CEO (“she”) is to make an investment decision, whereas the CFO (“he”) chooses the financing

of the investment. The project costs I and generates an uncertain return R, which equals either I + σ or

8

I − σ, each with probability 1/2, where σ ∈ (0, I] measures the return variability. If the CEO exerts effort,

she increases the expected return to R + ∆.12 Effort is costly, which is modeled as giving up a private benefit

B, similarly to Dewatripont and Tirole (1994) and Holmstrom and Tirole (1997, 1998).13 For simplicity, we

assume no discounting, and there are no other assets. As in the standard Tirole (2006) model, the CEO simply

maximizes the sum of expected net profit and private benefits.

The CFO’s job is to raise external financing to maximize firm value. He can either issue debt with a face

value of D or issue shares for a fraction γ of the firm. In both cases, external investors, who are risk neutral,

must break even in equilibrium. For tractability, we do not consider the possibility of issuing debt and equity

simultaneously.14 We assume that the CFO maximizes expected net returns.15

Both managers form expectations using their personal beliefs. Managers might be rational, or might exhibit

overconfidence. We define overconfidence as overoptimistic beliefs about the future cash flows accruing to the

firm. Specifically, an overconfident CEO believes that, by exerting effort, she increases cash flows by an amount

∆+ω. Similarly, an overconfident CFO believes that the cash flows increase by ∆+ω whenever the CEO exerts

effort. Both managers are aware of each other’s beliefs. When one manager is biased and the other is not, they

agree to disagree. We refer to both belief distortions as “overconfidence.” The common label is appropriate

in our context, despite the subtle conceptual differences between CEO and CFO overconfidence, as it captures

the same – late option exercise. However, for a CFO, late option exercise indicates an overestimation of the

future returns to the company at which he is employed, not necessarily an overestimation of his ability.

We focus the analysis on the parameter range where moral hazard affects both rational and overconfident

12 Note that, without effort, the expected net return is zero. This assumption merely reduces the number of cases, e.g., excludescases of severe financial constraints (very low E[R]) or cases where moral hazard becomes irrelevant to financing (very high E[R]).

13 See also Tirole (2010), Pagano and Volpin (2005), and Matsa (2010). In these papers, the private benefit is interpreted asthe benefit from working on pet projects (which reduces the expected revenue of the main project), as the personal benefit from“softer” management (less stress and confrontation), or simply as opportunity costs of managing the project diligently.

14 In principle, more complicated financing arrangements can be optimal. For example, in a setting with both moral hazardand risk-shifting incentives, Biais and Casamatta (1999) show that the optimal contract involves a combination of debt, equity,and, possibly, options. The restricted space of possible contracts allows us to guide the empirical tests of preferences for debt andequity financing in a simple setting.

15A way to formalize the CEO’s and the CFO’s incentives would be to assume that the CEO “hires” the CFO and pays hima fraction ε of the final cash flow, with the CEO obtaining a fraction 1 − ε. Given that the CFO’s problem is linear in the firmcash flow, the fraction ε could be made arbitrarily small, and for simplicity we ignore it in our exposition. Yet another possibilityis that the CFO gives some weight to the CEO’s well-being, which includes the private benefit B. In unreported results, we havemodeled the CFO as “fully committed to the CEO,” i.e., as maximizing the CEO’s expected utility including B, and the modeldelivers the exact same insights.

9

CEOs, ∆ > B ≥ ω. The first part of the double-inequality, ∆ > B, guarantees that the CEO’s effort is both

socially valuable and individually valuable to the (rational) CEO. The second part implies that the additional

return to effort an overconfident CEO mistakenly expects to obtain (ω) is bounded above by the private benefit

from shirking B. We show the robustness of our results to a broader parameter range in Appendix A.5.16

Investors anticipate correctly the true expected payoffs of the investment project. This modelling choice

embeds two assumptions. First, as in previous literature (see Malmendier and Tate (2005, 2008)), investors

do not share managers’ overly optimistic views. Second, investors rationally predict the effort the CEO will

put into the project in equilibrium. It is not necessary – though one possible interpretation – that investors

recognize managerial overconfidence and anticipate how it affects managerial behavior.17 For the empirical

results, it suffices that investors predict the managers’ effort choices correctly. For example, they may expect

managers to exert effort in bad states of the world because they perceive them to enjoy leisure less.

The timing is as follows. At t = 0, the CEO announces the planned investment, and the CFO chooses

between debt and equity financing. If funding is obtained, the profitability of the investment (I+σ or I−σ) is

revealed at t = 1, and the CEO decides at t = 2 whether to exert effort. At t = 3, the cash flow is realized and

investors are repaid. Figure 1 shows the full timeline. The dotted line on the left captures the model extension

from Section 2.5, where we analyze the possibility of endogenous pairing between CEO and CFO.

Before moving to the solution of the model, we emphasize that its timing and assumptions are flexible

enough to also accommodate other interpretations frequently offered in the literature. One common model

set-up, for example, revolves around liquidity shocks (cf. Holmstrom and Tirole (1997)): After receiving outside

financing, the manager has the option to either invest in a profitable project, or “shirk” and simply hold the

cash. Project profitability is fixed, but the cash flow available to investors is affected by a “liquidity shock”

known to the manager before making the effort choice. The liquidity shock might represent additional costs

of the initial investment, or a shortfall in returns. It might also be modeled as a mean-preserving spread as in

16 Broadly speaking, if the first part of the double-inequality does not hold, i.e., ∆ ≤ B, the rational CEO never exerts effort(except in the knife-edge case ∆ = B); and if the second part does not hold, i.e., B < ω, the optimal debt contract becomes morecomplicated, without generating new insights.

17 This assumption is supported by the evidence in Otto (2014) that shareholders recognize managerial optimism and adjustincentive contracts accordingly. It is also consistent with the evidence in Malmendier and Tate (2008) and Hirshleifer et al. (2012),who show that the option-exercise based measure of overconfidence is correlated with press portraits, suggesting that outsidersare able to identify overconfident managers.

10

Matsa (2010) (from whom we borrow our proxy for earnings volatility in Section 5.1). The key feature in all

of these cases is that a shock affects the firm’s net worth, which in turn affects its financing capacity.

2.2 CEO Overconfidence and Moral Hazard

Solving backward, we first analyze the effort decision of the CEO at t = 2, given the capital-structure choice of

the CFO at t = 0 and the state of the world drawn at t = 1. We denote the return the CEO expects to obtain

from exerting effort as ∆ + ωCEO with ωCEO = ω if she is overconfident, and ωCEO = 0 if she is rational. As is

standard in this type of model, we assume that the manager exerts effort rather than shirking when indifferent.

We have four incentive compatibility (IC) constraints regarding the CEO’s effort choice, one for each

financing choice and state of the world. For debt financing and in the good state, the CEO exerts effort if

max {0, I + σ + ∆ + ωCEO −D} ≥ max {0, I + σ −D}+B, (1)

where D is the face value of the debt. In the bad state, the IC for exerting effort under debt financing is

max {0, I − σ + ∆ + ωCEO −D} ≥ max {0, I − σ −D}+B. (2)

Under equity financing, the CEO retains a fraction (1− γ) of the payoff. In this case, the ICs for the good and

the bad state of the world, (1− γ)(I + σ+ ∆ + ωCEO) ≥ (1− γ)(I + σ) +B and (1− γ)(I − σ+ ∆ + ωCEO) ≥

(1− γ)(I − σ) +B, can both be simplified to

(1− γ)(∆ + ωCEO) ≥ B. (3)

2.3 CEO Overconfidence and the Cost of Debt

The CFO chooses between debt and equity at t = 0. We first derive the optimal debt contract conditional on

choosing debt and, in the next subsection, the optimal equity contract conditional on equity financing, and we

will then solve for the CFO’s choice between debt and equity.

We denote the return to the project in state S ∈ {Good,Bad} and with effort e ∈ {0, 1} as π(S, e). For

example, π(Good, 1) equals I + σ + ∆. Similarly, we denote the return expected by the CEO and the CFO,

given their beliefs, as πCEO(S, e) and πCFO(S, e). We express e as a function of S using the notation eS.

11

Conditional on debt financing, the CFO solves the following maximization program:

maxD

E[max {0, πCFO(S, eS)−D}] (4a)

uCEO(S,D, eS) ≥ uCEO(S,D, e′S) ∀S and eS 6= e′S (4b)

E[min {D, π(S, eS)}] ≥ I (4c)

where uCEO(S,D, eS) denotes the CEO’s utility in state S under a debt contract with face value D if she exerts

effort eS. The participation constraint in (4c) accounts for the possibilities that returns are larger than D, in

which case incumbent shareholders obtain the residual revenues, or lower than D, in which case debtholders

obtain all returns. We denote the face value of debt that solves the maximization problem given the CEO’s

belief ωCEO as D∗ω. We will see below that the optimal contract does not depend on the CFO’s beliefs.

Proposition 1 (Cost of Debt). The equilibrium cost of debt financing is lower for firms with

overconfident CEOs, and is independent of CFOs’ beliefs. Specifically, the cost is strictly lower for

intermediate ranges of return variability, with a face value of debt D∗ω = I for an overconfident CEO

and D∗0 = I+σ for a rational CEO. It is identical for sufficiently low variability (with D∗0 = D∗ω = I)

or high return variability (with D∗0 = D∗ω = I + σ).

Proof: See Appendix A.1.18

Proposition 1 delivers the prediction that overconfident CEOs work harder than rational CEOs when their

firm faces medium return volatility, holding constant investment opportunities, and are rewarded with better

financing terms.19 Intuitively, for small levels of ex-ante variability in returns, both types of CEOs exert high

effort regardless of the realized state of the world. Even in the bad state, payoffs are high enough to make effort

worthwhile for both types. For high levels of variability, instead, both types of CEOs shirk in the bad state.

Anticipating such behavior debtholders seek compensation in the good state of the world by imposing a higher

face value of debt. For moderate levels of variability, instead, the low payoff in the bad state deters a rational

CEO from working hard, but not an overconfident CEO, who overestimates the value she can generate.20

18 The thresholds for high, medium, and low variability are made precise in the proof.19 We obtain the same results if we reduce the role of the CFO to choosing debt or equity while the CEO rejects or accepts the

contract proposed by investors, i. e., if the contract maximizes the CEO’s rather than the CFO’s utility.20 In a model where managers also choose the investment level, this insight still holds when the resulting overinvestment problem

(Malmendier and Tate (2005), Goel and Thakor (2008)) is not “too severe” relative to the moral hazard problem.

12

We note that the result that overconfident managers may be more motivated to work hard is shared by

a number of models with biased agents, such as Benabou and Tirole (2002), Puri and Robinson (2007), and,

in the managerial context, Otto (2014). Hence, although our setting and predictions are specific, its main

message is common to a broader literature. We are, however, the first to translate, and test, the effect of

bias-induced effort into the cost of debt. The more subtle result that the effect is driven by medium-volatility

firms is, instead, distinctive of our modelling approach. What exactly constitutes a ‘medium range of volatility’

depends of course on the parametrization of our model, including the unknown traits (B,ω) of the CEO. In

the empirical analysis, we first use terciles and a number of different proxies for volatility to split the sample.

We then explore a wide range of alternative sample splits to test the robustness of the non-monotonicity result.

Turning to the optimal equity contract, conditional on equity financing, we derive the optimal contract in

a similar fashion in Appendix A.2. Here, the optimal contract either assigns a fraction γ∗ω = I/(I + ∆) to

outside investors, with the CEO exerting effort in both states of the world; or, if the moral hazard problem

is too severe, it assigns full ownership γ∗ω = 1, with the CEO not exerting effort in either state of the world.

As in the case of debt financing, overconfident CEOs also enjoy a lower cost of equity financing within certain

parameter ranges. However, the theoretical prediction varies with parameters B,∆, and I, which are hard to

pin down empirically, and is less robust, for example, to strategic reasons for equity issuance (signaling, market

timing). We will thus focus the empirical analysis of the cost of financing on the case of debt issuance.

2.4 CFO Overconfidence and the Choice between Debt and Equity

In order to solve for the CFO’s choice between the optimal debt contract (derived in the previous subsection)

and the optimal equity contract (derived in Appendix A.2), we compare his perceived expected utility under

the different financing scenarios. Since both a rational and an overconfident CFO correctly take the CEO’s

beliefs and their impact on the cost of debt and equity into account, even a rational CFO’s choice will be

affected by the CEO being overconfident. Proposition 2 summarizes the results:

Proposition 2 (Choice between Debt and Equity). An overconfident CFO uses (weakly) more

debt financing and less equity financing than a rational CFO, both under an overconfident and under

13

a rational CEO.

Proof: See Appendix A.3.

Intuitively, a biased CFO overestimates the return when the CEO puts effort into the project. For this

reason, he perceives external financing to be too costly. Under equity financing, this difference in opinion

matters for all states of the world; under debt financing it matters only for default states, which explains the

relative preference for debt. As spelled out more precisely in the proof, there are parameter ranges for which

both types of CFOs strictly prefer debt over equity; and there are parameter ranges where an overconfident

CFO strictly prefers debt over equity while a rational CFO does not. In the latter case, the overconfident CFO

uses more debt financing than a rational CFO (as long as the rational CFO does not always pick debt when

indifferent). This result is similar to the one in Malmendier et al. (2011), albeit applied to the CFO. Hence, our

main contribution does not lie in associating a preference for debt to overconfidence, but will be to distinguish,

in the empirical section, the roles of CEO and CFO biases as determinant of capital structure.

2.5 CEO Overconfidence and CFO Hiring

The CEO’s beliefs might also affect the selection of new CFOs. The recruiting of the CFO is a prerogative of

the board of directors. However, a large empirical literature documents the strong influence of the CEO on the

appointment of board members (Shivdasani and Yermack (1999); Cai et al. (2009); Fischer et al. (2009)), and

CEOs also control the selection of all other C-suite managers. In our simplified setting, we assign the CEO

sole discretion in replacing a CFO. For this part of the analysis, we add a period t = −1 in which the CEO

chooses the CFO.

Proposition 3 (CEO’s Hiring Decision).

An overconfident CEO (weakly) prefers to hire an overconfident CFO.

Proof: See Appendix A.4.

Although Proposition 3 may feel mechanical given our assumptions, it is not immediate since the CEO and

the CFO maximize different objective functions even when they share the same degree of bias. The reason for

the assortative matching result of Proposition 3 is that there is no disagreement regarding the CEO’s moral

14

hazard problem. All that matters for the financing choice of the CFO is the commonality or discrepancy

of beliefs. Since a rational CFO deviates from the preferred choices of the overconfident CEO (over some

parameter ranges), overconfident CEOs prefer the financial decision-making of overconfident CFOs on average,

and hence hire overconfident CFOs when given the opportunity.

2.6 Empirical Predictions

We summarize our findings formulated as three testable predictions:

Prediction 1. Overconfident CFOs are more likely to issue debt rather than equity when accessing external

financing, conditional on the CEO’s type.

Prediction 2. CEO overconfidence is associated with a lower average cost of debt. This effect is driven by

firms experiencing medium variability in their profits.

Prediction 3. A firm run by an overconfident CEO is more likely to hire an overconfident CFO.

3 Data

3.1 Overconfidence Measure

Measuring managerial overconfidence is a challenge to empirical researchers. The existing methodologies fall

into four categories: the option-based approach, the earnings-forecast-based approach, the survey-based ap-

proach, and the press-based approach. Option-based measures, first proposed by Malmendier and Tate (2005),

are by far most widely-used. The identification relies on individual choices and revealed beliefs: It infers man-

agers’ expectations about their companies from their personal investment choices. Managers who overestimate

future cash flows tend to overinvest their personal wealth in their companies and expect to personally benefit

from future stock-price increases. In particular, they do not diversify their stock-based compensation and

delay the exercise of executive stock options.21 This measure captures managers overestimating returns which

21 Another way to overinvest in the own company is to delay the sale of stock. Overconfident managers also exhibit thatbehavior, and even buy additional stock of their firms. Empirical research has relied more on option-based measures than on stockpurchases and sales as they raise fewer concerns about signaling to the market; cf. Malmendier and Tate (2008).

15

is exactly how overconfidence is defined in our model. Galasso and Simcoe (2011), Malmendier et al. (2011),

Otto (2014), and Hirshleifer et al. (2012) also adopt this measurement strategy.22 The earnings-forecast-based

approach, proposed by Otto (2014), infers overconfidence from overstated earnings forecasts. The survey-based

approach, developed by Ben-David et al. (2007) and Ben-David and Graham (2013), constructs CFO over-

confidence proxies based on miscalibrated stock-market forecasts of CFOs who participated in the Duke/CFO

Business Outlook survey.23 The media-based approach, employed by Malmendier and Tate (2008) and Hirsh-

leifer et al. (2012), constructs CEO overconfidence measures from the characterization of CEOs in the press.

In this paper, we follow the option-based approach, which is both most commonly employed in prior

literature and is generally most robust to alternative interpretations (see Malmendier and Tate (2015)). We

replicate and expand the Longholder proxy of Malmendier et al. (2011). In addition, we also replicate our

results using a continuous variant of the option measure proposed by Otto (2014).

The Longholder measure exploits the timing of option exercise to measure managerial overconfidence. It

is based on a benchmark model (Hall and Murphy (2002)), where the optimal exercise schedule depends on

individual wealth, risk aversion, and diversification. Given that executive stock options are not tradable, and

short-selling of company stock is prohibited, managers are highly exposed to the idiosyncratic risk of their

firms. Rational, risk-averse managers address their under-diversification by exercising in-the-money options

some time before expiration. Overconfident managers, who overestimate the future cash flows of their firms,

postpone exercising in order to tap expected future gains. Malmendier and Tate (2005) capture this insight

with a binary variable called Longholder. It indicates if a manager, at some point during their tenure, held an

option until the last year before expiration even though the option was at least 40% in-the-money.

In order to replicate the Longholder measure for a longer and more recent sample period, and for a broader

set of managers and firms, we build on the Longholder variant proposed by Malmendier et al. (2011). Their

proxy has the same definition as the original Longholder measure, but uses the Thomson insider filing data

set to identify option exercises of managers in public U.S. firms. We reconstruct the measure for our extended

sample period, and apply the measure to CFOs. The control group consists of managers who are also in the

22 Relatedly, Sen and Tumarkin (2015) measure overconfidence as the share retention rate after an option exercise.23 This behavioral bias reflects an underestimation of variance but is sometimes also called overconfidence. However, it does

not imply delayed option exercise. See Malmendier et al. (2011) (fn. 1) for a brief discussion.

16

Thomson data but do not meet the overconfidence criteria. We also use the data to construct a continuous

version of the Longholder measure proposed by Otto (2014). Here, we first calculate overconfidence indicators

for each option exercise, and then average all executive-specific indicators weighted by the number of shares

exercised. Details of the construction and replications of all estimations with the continuous measure are in

Appendix C.

The discrete and continuous measures are strongly correlated, with correlation coefficients of 41.9% for

CEOs and 46.5% for CFOs. The estimation results are also similar for our main specifications. They differ

only when we work with relatively small and selected samples. The differences may reflect the fact that the

dummy approach generates more variation than the continuous measure,24 or that the linearity implicit in

the continuous measure is an imperfect representation of the variation in overconfidence. In our context, we

choose the more widely used indicator version (in the main text) for a different and somewhat subtle reason: A

necessary condition for a CEO to be a Longholder is that she experiences at least one instance where options

are deeply in the money. In order to “score high” in terms of overconfidence under the continuous measure, the

manager needs to experience many such instances. This condition is very demanding in the more limited data

on CFOs, and likely to be met only for particularly successful companies. The indicator version avoids such

issues of selection or misattribution. At the same time, we acknowledge the appeal of a continuous measure

with its finer distinction, and replicate all estimations with the continuous measure in Appendix C.

We use the Thomson insider filing data to construct the updated Longholder measures for both CEOs and

CFOs. Thomson collects data from Forms 3, 4, and 5, which insiders report to the SEC. The data consists of

two data sets called “Table 1” (Stock Transactions) and “Table 2” (Derivative Transactions). We extract the

option exercise data from the “Table 2” data, which contains information from Form 4. (Changes in ownership

must be reported to the SEC within two business days on Form 4.) These transactions data are available since

1996. However, as Longholder is constructed as a permanent characteristic, we can include the years 1992-1995

for those companies into our sample that had managers for which we can obtain transactions data in Form 4.

We keep only records with Thomson cleanse indicators R, H, and C (very high degree of confidence in data

24 For example, the standard deviations of the Longholder CEO and Longholder CFO dummies are 0.46 and 0.49, respectively,in our largest sample, but only 0.017 and 0.07 for the continuous measure.

17

accuracy and reasonableness) or Thomson cleanse indicators L and I (reasonably high degree of confidence).

Following prior literature (e.g., Lakonishok and Lee (2001)), we drop records that are amendments to previous

records and records with obvious errors, such as an indicated maturity date that is earlier than the exercise

date, or options with missing exercise date. We also remove outliers with exercise prices below $0.1 or above

$1000. We calculate the percentage-in-the-money for each option using stock price data from CRSP.

We obtain tenure as well as stock and option holdings of the CEOs and CFOs from ExecuComp. This step

limits our sample to the intersection of ExecuComp and Thomson data, i. e., a subset of the S&P 1500 small-,

medium-, and large-cap firms from 1992-2015. We use CUSIPs to merge the firm-level information in Thomson

and Compustat/ExecuComp, and employ a conservative fuzzy algorithm to link the names of the executives

in the two data sets. We verify manually the accuracy of each match. In a few cases a firm has more than

one CEO or CFO listed in ExecuComp. In these instances, we manually check the 10-K forms on the SEC

website25 and identified the executive who held the relevant position at the end of the fiscal year.

An empirical issue with the CFO data is the significantly lower number of transactions available to construct

the overconfidence proxy. CFOs typically receive smaller option grants than CEOs, and are also covered less in

ExecuComp. This introduces measurement error when we categorize a CFO as non-overconfident. To address

this problem and ensure that we capture a systematic behavior, we keep only managers for which we observe

at least ten transactions. However, as discussed in more detail in Section 3.3 and shown in Appendix-Figure

C.1, our estimates remain very stable when we alter the filter requirement.

Table 1 summarizes the data construction. Of the 8,054 CEOs and 7,402 CFOs in ExecuComp, about 20%

(1,623 CEOs and 1,246 CFOs) are also recorded in Thomson and report at least ten transactions, corresponding

to 5,810 firm-years. After dropping financial, utilities, and firms with missing manager or firm controls, the

final sample consists of 4,581 firm-years.

25 See http://www.sec.gov/edgar.shtml. The Edgar database contains 10-K forms starting in 1994. For some earlier cases wecannot recover the information and exclude those observations.

18

3.2 Alternative Interpretations

Before turning to the remaining data construction, we address possible alternative interpretations of the

Longholder measure and their implications for the results of this paper.

Procrastination. The Longholder overconfidence measure captures a persistent tendency of managers

to delay option exercise. One might be concerned that such behavior indicates inertia or procrastination.

We find, however, that 74% of overconfident CEOs and 69% of overconfident CFOs conduct other portfolio

transactions one year prior to the year when their options expire, which is inconsistent with the interpretation

that Longholders would persistently delay managing their personal portfolios.

Insider information. Managers may choose to hold exercisable options because they have positive inside

information about future returns. One issue with this alternative explanation is that inside information should,

by definition, be transitory rather than persistent, but Longholders persistently hold exercisable options for

several years. A second issue with the inside-information interpretation is that insiders should earn positive

abnormal returns from holding options until expiration. While we cannot calculate expected returns from an

ex-ante perspective, we can calculate the actual returns of Longholder CEOs and CFOs from holding options

that were at least 40% in-the-money (“Longheld” transactions) until their expiration. We compare these

actual returns to hypothetical returns from exercising these options 1, 2, 3, or 4 years earlier and investing the

proceeds in the S&P 500 Index until the options were actually exercised. We find that, depending on the horizon

chosen, approximately 45-48% of the “Longheld” transactions do not earn positive abnormal returns. We then

re-estimate our regression model on the subset of Longholders who lose money by holding their options. The

new estimates either confirm or even strengthen the results, whenever the sample is large enough to separately

estimate “winner” and “loser” Longholder variables. The same has been found in previous research employing

Longholder-type measures; see, e.g., Malmendier and Tate (2008).

Signaling. One might argue that Longholders intend to signal to the capital market that their firms have

better prospects than other, similar firms. However, as in the discussion of insider trading, the persistence

of Longholders’ behavior is hard to reconcile with the signaling interpretation. Private positive information

revealed by a manager through option exercise should be quickly incorporated into prices (as long as the

19

manager is rational and, hence, credible) and thus give rise to an occasional, rather than persistent, behavior.

Moreover, all estimations control for the number of vested options held by the manager (standardized by the

total number of shares outstanding) to account for the possibility of signaling via option holdings.

Risk tolerance. The Longholder overconfidence measure captures a habitual tendency of managers to

hold company risk. One might be concerned that risk-tolerant or risk-seeking managers prefer to hold exer-

cisable options longer, and therefore appear to be overconfident under the Longholder measure. However, the

executive’s willingness to hold risk should be reflected in the riskiness of the overall portfolio of options or

shares she is willing to hold at any given point in time, not the timing of exercise per se. Hence, we control

for vested options and shares held in our estimations.

Agency problems. Another alternative interpretation is that, being more incentivized, option-holding

managers are more willing to act in the interest of (existing) shareholders. However, in all of our regressions,

we control for both the shares and the vested options owned by managers. Moreover, the observed differences

in the behavior of Longholders, compared to managers who diversify their holdings, are not easily interpreted

as shareholder-value maximizing. By increasing leverage, Longholders likely reduce the cash flow available to

shareholders. This behavior might be costly to shareholders if it increases default probability and if there are

non-negligible bankruptcy costs.

Firm performance. Another concern is a potential mechanical correlation of the Longholder measure with

past performance. Given the construction of the proxy, an executive cannot be identified as overconfident unless

her firm’s stock has appreciated by at least 40%. Therefore, one may worry that, in our empirical analysis,

overconfident managers are simply those running particularly successful firms. To address this confound, we

compute, for each firm, the buy-and-hold return over the previous 1, 2, 3, 4, and 10 years and test whether they

are systematically correlated with the overconfidence measures. We find that the correlations of the Longholder

dummies with lagged buy-and-hold returns are small and often negative. For example, at the ten-year horizon,

which is the most relevant horizon for our analysis, the two correlation coefficients are very small in absolute

value and of opposite signs, positive for Longholder CEOs (0.024) and negative for the Longholder CFOs

(-0.009). This is at odds with the idea that our measures capture a common pattern of past performance.

As a second way to address concerns about links with past performance, we re-run our analysis on the

20

subsample of firms that have appreciated by more than 40% in the previous ten years. This subsample

selection is quite restrictive—not because of the 40% requirement; but because it excludes all firms that, in a

given year, have less than ten years of past data.26 Despite the significant loss of sample size and power, we

replicate our estimations on this subsample. Our main results are qualitatively and quantitatively unaffected,

except in instances where we work with a very small sample and where the data offer limited variation due to

our empirical strategy (cf. Table 4). We include examples of estimations of a CFO effect (debt issues) and a

CEO effect (net interest) in Appendix-Tables C.8 and C.9.

Mismeasurement. The Longholder proxy draws a simple dichotomous distinction between overconfident

and rational managers. It may thus be susceptible to mismeasurement in at least two ways. First, it is sensitive

to errors in the Thomson Reuters data (e. g., in the grant or expiration dates of the options). Second, it does

not distinguish between managers who display a stronger or weaker tendency to exercise options late.

The continuous version of the Longholder measure developed by Otto (2014) is unlikely to be affected

by occasional errors in the Thomson database, and allows us to distinguish more finely different degrees of

overconfidence. As reported in Appendix C, we obtain largely similar results when following this approach.27

Tax advantages. Another possible concern are tax reasons. The informal argument, also discussed in

McDonald (2005), goes as follows: When exercising a stock option, executives pay ordinary income tax. Upon

sale of the underlying stock, they pay only capital gains tax, which is typically lower. Early exercise may thus

be optimal if the stock price is expected to rise, and late exercisers may be those who predict poor performance.

If option owners are on average correct, late exercise would be correlated with poor future performance.

The discussion of signaling and insider information has already emphasized that the Longholder proxy

captures persistent behavior. While a manager may be more or less pessimistic about the firm’s prospects at

a given time, a standard model cannot explain a systematic pattern of optimistic (or pessimistic) expectations

26 The ten-year restriction reduces the sample by about 26%, but only 18% of the remaining firm-years feature returns lowerthan 40% over the previous ten years.

27 To summarize briefly, results using the alternative measures are similar for the analyses of Debt Issuance using Compustatand CFO Hiring (Tables C.2 and C.6 in Appendix C); qualitatively similar but slightly weaker statistically for the Interest Ratesregressions (Table C.5); statistically stronger for the Leverage regressions (Table C.4); and inconsistent only for the regressionswhich adopt the “Financing Deficit” approach (Table C.3). Also, quantitatively the variation explained under the dummy measureand the variation explained under the continuous measure are of the same order of magnitude. For example, a one-standard-deviation increase in CFO overconfidence using the continuous measure increases the odds ratio of issuing debt by 15.8% (TableC.2), which is in line with our results of Table 3.

21

and resulting option exercise behavior. Moreover, stocks owned by Longholders do not appear to under- or

outperform the market in the long run. Finally, and perhaps most importantly, the intuitive argument sketched

above may be appealing but, as McDonald (2005) shows, its logic is not correct. In a rigorous framework where a

manager does not borrow money to pay income taxes, accelerating the option exercise is not optimal in general.

Other concerns. A large literature has discussed many other potential interpretations of the Longholder

measure. Malmendier and Tate (2005) address concerns about industry selection, proxying for observable firm

or manager characteristics, in addition to an in-depth analysis of inside information, signaling, risk tolerance,

taxes, and procrastination. Malmendier and Tate (2008) also discuss concerns about CEO preference, past

performance, market inefficiency, board pressure, corporate governance, taxes, and dividends. Malmendier et al.

(2011) address concerns about dilution and more aspects of inside information, signaling, and risk tolerance.

Hence, while any empirical estimates using option-based overconfidence measures must be subjected to

scrutiny as they are not the result of exogeneous variation, the leading alternative interpretations appear to

be addressed either in the details of the construction of the measure, or in the empirical results.

3.3 Other variables

Our analysis requires a broad array of firm-level financial variables as well as other firm and industry char-

acteristics. We retrieve these variables from Compustat, excluding financial firms and regulated utilities (SIC

codes 6000-6999 and 4900-4999) for the usual concerns about the lack of comparability in the accounting data.

Below, we describe briefly the main variables of interest. Additional details are in Appendix B.

The key variables for our analysis of financial policies are Net Debt Issues and Net Financing Deficit.

Following Malmendier et al. (2011) and other prior literature (cf., among many others, Shroff (2015)), we

construct Net Debt Issues as long-term debt issues minus long-term debt reductions. Net Financing Deficit is

cash dividends plus investment plus the change in working capital minus cash flow after interest and taxes.

Both variables are normalized by assets at the beginning of the year.

Standard firm-level control variables include Q (assets plus market value of equity (price times common

shares outstanding) minus common equity and balance sheet deferred taxes and investment tax credit, all

22

divided by assets), profitability (operating income before depreciation, normalized by assets at the beginning

of the year), tangibility (property, plants and equipment normalized by assets at the beginning of the year),

size (natural logarithm of sales), book leverage (sum of debt in current liabilities and long-term debt, divided

by the sum of debt in current liabilities, long-term debt, and common equity), and annual changes in these

variables. We combine manager-level variables with firm-level variables to form the whole sample, a panel of

679 S&P 1500 firms from 1992 to 2015, corresponding to 4,581 firm-years indicated in Table 1.

Table 2 reports summary statistics for firm-level variables in Panel A, and for CEO- and CFO-specific

variables in Panel B. Each panel contains separate tables for the different (sub-)samples used in each analysis.

Not surprisingly, the typical company in our data set is large relative to the Compustat universe. The average

revenues in our overall sample (the data used in the Financial Deficit analysis of Table 4) amount to $5.7 billion,

relative to a mean of $2.5 billion for the full Compustat data set over the same time period. Our companies

also tend to have slightly lower book leverage (28.9% versus 31.5%) and significantly higher profitability (18.5%

versus 7.0%). Relative to the ExecuComp data, of which our data is a subset, the differences are much less

pronounced. The respective figures are $4.7 billion, 36.2%, and 13.3%. Hence, our sample appears to be fairly

similar to those studied in past empirical work on executive compensation.

Panel B of Table 2 reveals that, on average, CEOs tend to own significantly more stock of their companies

than CFOs (1.81% versus 0.12% in the sample used in Tables 4 and 5). The difference is somewhat less

pronounced for vested options (1.04% versus 0.26%). These figures are comparable to those we obtain when

analyzing the full ExecuComp dataset.28 We have also analyzed managerial controls separately for the full

sample and for overconfident managers, and find that they tend to have fairly similar equity incentives.29

For completeness, Appendix-Table C.1 reports the descriptive statistics for our largest sample, used in

Tables 4 and 5 of the paper, split by the four possible combinations of executives’ biases (both executives

rational, both overconfident, rational CEO and overconfident CFO, overconfident CEO and rational CFO).

28 The average stock ownership in ExecuComp is 2.43% for CEOs and 0.15% for CFOs; the number of vested options scaled bynumber of shares is 0.73% for CEOs and 0.15% for CFOs. Thus, executives in our sample have similarly powered equity incentives,with a slight tilt toward options rather than stock, possibly reflecting the merge with option exercise data from Thomson.

29 We have also included, in unreported tests, gender and age as additional managerial controls. Gender is available for allmanagers but exhibits very little variation, with only 2% of CEOs and 8% of CFOs being women. It has no effects on our estimates.When including executives’ ages as control variables we also find very similar results; given that age is missing for about 10% ofthe CFOs we chose to omit it and work with a more comprehensive sample.

23

Compared to the samples used in Malmendier and Tate (2005, 2008) and Galasso and Simcoe (2011), the

Thomson and ExecuComp-based data sets in Malmendier et al. (2011) and Hirshleifer et al. (2012), and also

compared to the survey sample of Ben-David and Graham (2013), our sample differs in three ways: First, it

extends to a more recent time period. Second, it considers small and medium firms in addition to large firms.

And third, it includes overconfidence measures for the CFO. The last difference is key in that we aim to fill a

gap in the existing literature by estimating the effects of CEO and CFO overconfidence separately and jointly.

These differences in sample composition are important to recognize in order to understand differences in

the observed frequency of overconfident managers. In our sample, 66.5-69.8% of CEOs and 52.8-57.5% of

CFOs are Longholders. These frequencies are two to three times as high as in the first wave of overconfidence

research, which used option exercise data from the 1980s to mid-1990s, but in line with the more recent wave

of research, which also uses the more recent option-exercise data, e. g., Malmendier and Tate (2015). An

interesting observation is that, without the restriction to managers with at least 10 transactions, the relative

frequencies of firm-years with overconfident managers drop to 60% for CEOs and 43% for CFOs. Thus, the

restriction increases the percentage of overconfident CFOs considerably more than that of overconfident CEOs.

Because CFOs’ options packages are in practice much smaller than those of CEOs (see Table 2, Panel B), this

observation cautions that managers with fewer opportunities to trade options are less likely to be classified as

overconfident. Hence, a sample restriction to managers with similar transaction frequencies might generally be

in order, especially when looking at CFOs or other managers that are less well covered than CEOs.

At the same time, we recognize a trade-off in using such a filter. While our proxy becomes more reliable, we

are constrained to a smaller sample, which may suffer from different selection problems. Thus, we have re-done

our analysis relaxing this restriction, requiring 1, 2, ..., 9 transactions recorded. (For example, when we require

a single transaction registered in Thomson, we have 10,184 firm-years, more than twice the size of the sample

with the 10 transactions requirement). We find that our results are largely unaffected.30 For reference, we plot

our main coefficient of interest in each regression versus the minimum number of transactions required for each

executives in Appendix Figure C.1. (We show, for brevity, only the coefficients from the most conservative

specifications.) Economic magnitudes are generally larger with stricter requirements when we focus on CFOs,

30 The only exception is given by the leverage regressions of Section 4.3, where we use firm-fixed effects.

24

and are roughly independent from such restrictions when we look at the debt regressions (where we focus on

CEOs). Hence, as expected, measurement error in the construction of the Longholder proxy is more of an issue

with CFOs.

Finally, we merge the ExecuComp-Compustat with the Dealscan data on syndicated loans to test our

predictions on the cost of debt. Dealscan provides detailed information on the pricing, type, maturity, and

size of loans. The coverage is typically limited to large and medium size firms, which are the main focus of

our analysis. We merge this data with the quarterly Compustat file, using the mapping provided by Chava

and Roberts (2008).31 Our outcome of interest is the amount the borrower pays in basis points over the

London Interbank Offered Rate, a variable called allindrawn in Dealscan. Our main specification uses 1,651

observations (408 different firms). We discuss the main control variables in more detail in Section 5.

4 Overconfidence and Financing Choices

Prediction 1 of our model is that overconfident CFOs exhibit a preference for debt over equity, conditional

on accessing the market for external financing. We test both the influence of the CFO, as predicted by the

theoretical model, and the influence of the CEO, whose overconfidence was the focus in prior literature.32

We use three different empirical approaches. First, we focus on firms that access external financing in a

given year, and ask whether they are more likely to issue debt than equity when lead by overconfident managers

(Section 4.1). We estimate the corresponding logit models both on Compustat and SDC, with the latter being

relegated to Appendix C for brevity. These analyses restrict the sample to firms that, in a given year, issue either

debt or equity. Hence, we cannot include firm fixed effects to control for time-invariant firm characteristics for

lack of sufficient variation. Under a second and third approach, instead, we use our full sample and control

for firm fixed effects. The second approach (Section 4.2) employs the standard ‘financing deficit framework’

of Shyam-Sunder and Myers (1999), also used in Malmendier et al. (2011). The third approach (Section 4.3)

31 The crosswalk is available up to 2012 at finance.wharton.upenn.edu/ mrroberts/styled-9/styled-12/index.html.32 It would be interesting to test whether the influence of CFO overconfidence diminishes when a CEO is less likely to fully

delegate financing decisions, e. g., when the CEO has experience as CFO or when the CFO has been appointed only recently and isrelatively unexperienced. Unfortunately, the resulting CEO sample is small and the CFO appointment information often missing.

25

extends the analysis to ask whether the influence of managerial characteristics on the flow of financing is

strong and persistent enough to affect firms’ capital structure, above and beyond the influence of permanent

firm characteristics. If so, firms run by overconfident executives should be systematically more leveraged.

4.1 Debt Issues

We first test whether overconfident managers are more likely to issue debt than equity using Compustat data.

As implied by the model, we condition the regression analysis on firms accessing external capital, controlling

for potential differences in the baseline frequencies of debt and equity issues by overconfident managers and

their rational peers. Thus, the regression sample only includes observations with either positive net debt issues

or positive net equity issues. In total, we have 2,939 firm-years with external financing (635 firms).

We estimate the following logit model:

Pr(NDIi,t|LTCEOi,t,LTCFOI,t, Xi,t, δt) = G(β1LTCEOi,t + β2LTCFOi,t +X ′i,tB + δt + εi,t), (5)

where G is the cumulative logistic distribution function, and the subindices i,t indicate year t in which company

i accessed external financing. The dependent variable NDIi,t is an indicator of firm i issuing positive net debt in

year t. As defined in Section 3.3, net debt issues are long-term debt issues minus long-term debt reductions.33

LTCEOi,t and LTCFOi,t represent the Longholder measures for the CEO and the CFO, respectively. Xi,t is

the vector of firm- and manager-level control variables for firm i in year t. Firm-level control variables are the

traditional determinants of capital structure—book leverage, log(Sales), profitability, Q, and tangibility—, and

also include two-digit SIC industry fixed effects, following the specification of Ben-David and Graham (2013).

Manager-level control variables are stock ownership and vested options, to control for the incentive effect of

stock-based executive compensation. In addition, we include a vector of year fixed effects. Standard errors are

adjusted for firm-level clustering, here and in all the estimations that follow. We note that the fixed effects

do not cause incidental parameter problems in this specification.34 Coefficient estimates are transformed to

33 Alternatively, we construct debt issues as change in total assets minus net equity issues and retained earnings, followingBaker and Wurgler (2002). We generally find very similar results, in particular if we include the full set of control variables.

34 The incidental parameters problem arises in panel estimations if, with increasing sample size, the number of fixed-effectparameters also grows, so that the coefficients are not estimated consistently. This does not apply to industry fixed-effects (Besterand Hansen (2016)). Nevertheless, we have checked that our results do not change if we estimate a linear probability model or a

26

indicate, for a unit increase in each independent variable, the expected change in the log odds of issuing debt.

Table 3 reports the results. We start by only including the CEO overconfidence proxy (columns 1 and 2),

replicating the analyses of prior literature. We then use the CFO instead of the CEO measure (columns 3

and 4), which captures the predictions of our model. Finally, we include both overconfidence measures jointly

(columns 5 to 7). The joint analyses test which managerial trait predicts a more pronounced pecking-order

preference, and whether the separately estimated overconfidence coefficients are robust.

In the baseline estimations with only the CEO overconfidence proxy we estimate a small positive and

insignificant log odds ratio when we only control for industry effects (column 1); the coefficient becomes

negative, and even smaller in magnitude, when we include the whole range of firm- and manager-level controls

as well as year dummies, which remove cyclical effects of debt issues (column 2). The coefficients of the

firm-level control variables are generally similar to those in prior capital-structure literature. Size is positively

related to the likelihood of debt issues, possibly reflecting easier access to bank loans or bond markets for

larger firms with sufficient collateral. Profitability and tangibility also have the expected positive sign, but are

not statistically significant predictors of debt issuance. Q is negatively related to debt issues, although not

significantly. Most importantly, the inclusion of control variables does not alter the lack of explanatory power

of the CEO overconfidence proxy. This result is consistent with Malmendier et al. (2011), who find a positive

but insignificant effect of CEO overconfidence on debt choice and a significantly negative effect on equity issues.

We will see somewhat stronger differences in results relative to prior literature in the analyses predicting the

Financing Deficit and Leverage, and discuss the likely underlying causes.

In columns 3 and 4, we turn to the actual model prediction and replace the CEO overconfidence measure

with the CFO overconfidence measure. For the baseline regression with only industry controls, the estimated

coefficient of the CFO overconfidence measure is large and significant at the 1% level. It indicates that the

odds of debt issues are 45% higher for overconfident than for rational CFOs. This finding remains unaffected

when we control for CFO-level variables, firm-level variables, industry dummies and year dummies in column

4. Note that the stability of the coefficient estimate also addresses concerns about potential confounds related

conditional logit model. We also get similar estimates with a coarser industry classification (Fama-French 12 industries). Theseremarks also apply to the results of Appendix-Table C.7, where we adopt the same empirical strategy for the SDC data.

27

to executives’ risk tolerance (see Section 3). If risk tolerance, rather than overconfident beliefs, induced a

Longholder to issue more debt and raise default risk, the explanatory power of Longholder CFO should decline

once we include both book leverage (as a measure for firm-level risk) and vested options (as a proxy for a

manager’s willingness to hold risk). However, the coefficient is unaffected by the inclusion of these variables.

In columns 5 to 7, we include both CEO and CFO overconfidence, first in the baseline regression, then

adding managerial controls, and finally including the full set of controls.35 These specifications test whether the

finding of a significant CFO effect is robust to the inclusion of the corresponding CEO controls. We find that,

while the coefficient on CEO overconfidence remains insignificant, CFO overconfidence retains its economic

and statistical magnitude. Further, the coefficients on CEO and CFO overconfidence are statistically different

at the 10%, 5%, and 1% level in columns 5, 6, and 7 respectively. The Longholder CFO coefficient of 0.437 in

the estimation with the full set of controls (column 7) implies that an overconfident CFO is 55% more likely

than a rational CFO to issue debt, conditional on accessing external markets. The Pseudo R2 is 17%, closely in

line with previous capital structure fixed-effect regressions on debt issues and previous literature on managerial

overconfidence. Note that the partial R2 of the overconfidence proxy is naturally low in an industry fixed-effects

regression,36 indicating other drivers of capital structure. We will detect a larger role in the estimations using

the full sample below. Here, the key insight is that we have detected a significant determinant, corroborating

the role of overconfident beliefs and disentangling the role of CFOs and CEOs.

For robustness, we also explore to what extent mismeasurement in growth opportunities migh contribute

to the relatively low R2 in some of our regressions. We use the minimum distance estimator developed by

Erickson, Jiang, and Whited (2014) to control for measurement error in Q. We replicate all the tests of this

paper and find that, while the explanatory power of Q does increase, our results are mostly unaffected.

As another robustness check, we have re-estimated the model in (5) using the SDC data on equity and

bond issues by US corporations, following Malmendier et al. (2011). The advantage of the SDC data is that it

35 The significant negative sign on CFO Vested Options may seem surprising but is consistent with the coefficient on CEOVested Options in Table 2 Panel B of Malmendier et al. (2011) which finds CEO Vested Options to be significantly positivelyrelated to the likelihood of choosing equity. They suggest this result is primarily driven by outliers. It may also reflect unobserveddeterminants. Importantly, column 5 demonstrates that our results are not driven by the inclusion of this variable.

36 For example, we find that the R2 increases by .55% relative to the R2 if CFO overconfidence is not included in the regression,16.25%, which is equivalent to calculating the partial R2 via the partial correlation.

28

identifies the timing of issuances more precisely, relative to the (noisier) accounting data from Compustat. Its

disadvantage is that it misses increases or decreases in firms’ external financing that are not recorded as new

issues in SDC, and that the sample size is much smaller. More details regarding the sample construction are

in Appendix C. As Appendix-Table C.7 shows, we obtain similar results.

4.2 Financing Deficit and Managerial Overconfidence

We now turn to our second approach of testing Prediction 1, using the standard financing deficit framework of

Shyam-Sunder and Myers (1999). This framework allows us to analyze whether, for a given need of external

funding, managers display a preference for debt over equity. We examine the impact of managerial overcon-

fidence on the association between the net financing deficit and the choice of external financing, as done in

Malmendier et al. (2011) for the CEO. The estimation framework allows for overconfident managers and their

rational peers to have different baseline needs for external financing. Another advantage of this approach is

that it utilizes all firm-years, resulting in a larger sample.

We estimate OLS regressions using the following equation:

Di,t = β1FDi,t + β2LTCEOi,t + β3LTCFOi,t + β4FDi,tLTCEOi,t

β5FDi,tLTCFOi,t +X ′i,tB1 + FDi,tX′i,tB2 + θi + δt + εi,t,

(6)

where Di,t is Net Debt Issues and FDi,t is the Net Financing Deficit, which measures the amount of external

financing needed in a given year. LTCEOi,t and LTCFOi,t are our measures of managerial overconfidence, and

Xi,t is a set of manager- and firm-level control variables including executive stock and vested options holdings,

changes in Q, profitability, tangibility, and size. In the most conservative specifications, we also interact our

vector of controls with the Financing Deficit. We note that the coefficients on the control variables generally

show the expected signs (not shown for brevity).37 We also include firm and year fixed-effects. The coefficients

of interest β4 and β5 measure the effects of CEO and CFO overconfidence, respectively, on debt financing,

conditional on the financing deficit. If, for given financing needs, overconfident CFOs issued disproportionately

more debt than unbiased managers, as predicted by our model, we would estimate β5 to be positive.

37 For example, Q is negatively related to debt issuance, whereas tangibility and size exhibit a positive association. (All variablesare in first differences.)

29

We start again from the relationship between CEO overconfidence and financing, which has been the focus

of prior research, before turning to CFO biases, which our model predictions pertain to. The baseline regression

in column 1 of Table 4 includes only the CEO overconfidence measure, its interaction with the net financing

deficit, and firm fixed effects. Column 2 adds the full set of control variables, including CEO stock and option

holdings, firm-level variables, and year fixed-effects. In column 3, we further add the interactions between the

net financing deficit and the control variables. (In columns 3, 6, and 9, FD is interacted with year dummies, so

the stand-alone variable is redundant and omitted from the table.) Across all three specifications, we find little

evidence for a role of CEO overconfidence, consistent with the results from the debt issuance regressions above.

The coefficients of CEO overconfidence interacted with net financing deficit are positive but insignificant, except

in column 3, where the coefficient is equal to 0.164, and is significant at the 5% level. These results imply a

weaker role of CEO overconfidence than prior literature (cf. Malmendier et al. (2011)), even before accounting

for the role of the CFO. The differences are possibly due to the more recent sample and – in light of our

CFO results – more specialization and delegation of financing decisions to the CFO. In Section 6, we discuss

in more detail how the CEO’s hiring of like-minded CFOs may help to explain the lower significance of CEO

overconfidence.

In columns 4 to 6, we employ the specifications from columns 1 to 3 but replace the CEO with the CFO

overconfidence measure. CFO overconfidence increases the sensitivity of net debt issues to the net financing

deficit significantly. The coefficient estimates of the interaction with net financing deficit lie between 0.179 and

0.243. These results corroborate our finding that CFO biases influence a firm’s tilt towards debt financing.

Finally, we include CEO and CFO overconfidence measures jointly (columns 7-9). The results remain very

similar to the separate estimations. The estimated effect of CFO overconfidence on the sensitivity of net debt

issues to the net financing deficit ranges from 0.166 to 0.247, and is significant at the 1% or 5% level. The

effect of CEO overconfidence remains small and insignificant. The CEO and CFO coefficients are statistically

different at the 10% level in the most conservative specification of column 9. The estimated effect of CFO

overconfidence is also quantitatively important. Consider the stand-alone coefficient of 0.094 on the financing

deficit in column 8. This sensitivity more than triples for overconfident CFOs, to 0.302 (0.094 + 0.208).

Moreover, the statistical significance of the coefficient tends to grow in the most demanding specifications,

30

in which the control variables are interacted with the financing deficit (columns 6 and 9), suggesting that

Longholder CFO is not simply picking up variation associated with other known predictors of debt issuance.

We also note that the variation in net debt issues explained by CFO overconfidence is substantial. In column

9, the R2 rises from 30.8% if the interaction between Longholder CFO and the net financing deficit is excluded

(unreported) to 49% when we include it as explanatory variable.

Taking the results from the three estimations of overconfidence on debt issuance together, CFO overconfi-

dence emerges as a statistically and economically significant determinant while CEO overconfidence exerts at

most marginal influence. These findings are consistent with Prediction 1 of our model.

4.3 Leverage and Managerial Overconfidence

Given the magnitude of our estimates so far, it is conceivable that the effect of managerial overconfidence might

even translate into a measurable impact on firms’ capital structure. As overconfident CFOs tend to prefer debt

over equity issuances, their companies should display, on average, higher leverage. Note that this implication

holds only if the overconfidence-induced bias towards debt is persistent and strong enough to dominate other

determinants of leverage, e.g., the well-documented persistence of past leverage ratios.

To investigate this question, we estimate the following specification, following the empirical strategy in

Bertrand and Schoar (2003) and Malmendier et al. (2011):

Leveragei,t = β1LTCEOi,t + β2LTCFOi,t +X ′i,tB + θi + δt + εi,t. (7)

LTCFOi,t and LTCEOi,t are the usual Longholder proxies for managerial overconfidence, Xi,t is a vector of

control variables, θi are firm fixed effects, and δt are year dummies. After controlling for firm fixed-effects,

the identifying variation comes from firms that switch from an unbiased to an overconfident manager, and

vice versa. Our dependent variable is market leverage, expressed as the ratio of long-term debt plus debt in

current liabilities over the market value of assets, i.e., over market capitalization (price times common shares

outstanding) plus the value of debt from the numerator, all multiplied by 100. This estimation uses again the

full sample.38

38 We lose 24 observations relative to the specification in Table 4 because either long-term debt or short-term debt is missing.

31

Table 5 reports the results. In column 1, we include only Longholder CEO, plus firm and year dummies. The

coefficient estimate for CEO overconfidence is positive but very small and insignificant. Even if the coefficient

of 1.485 were significant, it would imply that switching from a non-overconfident to an overconfident CEO

induces an increase in leverage by slightly more than 1 percentage point, relative to a sample mean of 14.57

(and standard deviation of 15.36). The coefficient estimate is further reduced, and remains insignificant, when

control variables are included (column 2). All the firm level control variables, on the other hand, have the

expected sign. Larger firms with higher tangibility are more levered, whereas profitability and Q are negatively

related to leverage. We do not find any association with managerial controls (shares and vested options owned).

Hence, we estimate a weaker role of CEO overconfidence than prior literature (cf. Malmendier et al. (2011)),

even before accounting for the role of the CFO. As mentioned before this indicates that, in addition to our new

insight—the confound with the CFO’s role and biases—we are also encountering slight sample differences.

Turning to the CFO effect, in columns 3 and 4, we estimate a strong and sizeable positive association

with market leverage. It makes little difference whether or not we include control variables. In column 4, the

coefficient is 3.678 (with a t-statistic of 2.815). When we consider both managerial biases jointly, in columns

5 and 6, the effect of CEO overconfidence vanishes further, while the coefficient estimate on Longholder CFO

becomes slightly larger and more precisely estimated, e.g., 3.800 with a t-statistic of 2.904 in the specification

with the full slate of controls (column 6). Among the managerial controls, CFO stock ownership is negatively

related to leverage, perhaps because risk aversion induces CFOs to adopt more conservative financial policies

when their wealth is heavily invested in their company. To further probe the robustness of this result, we also

add controls for financing deficit (in column 7) and lagged one-year returns (in column 8).39 The coefficient

on Net Financing Deficit is positive, giving support to traditional pecking-order models of corporate financing

(Shyam-Sunder and Myers (1999)). The coefficient on past returns is negative, likely capturing both market

timing reasons (see, e.g., Welch (2004)) and a mechanical effect: past high returns lower market leverage simply

because they increase the denominator. In all cases, our coefficient of interest is unaffected.

In terms of fit, the inclusion of Longholder CFO increases the R2 by about half a percentage point as we

can see, for example, comparing columns 2 and 6. This number is not large, but not negligible either, given

39The coefficients on CEO and CFO overconfidence are statistically different at the 10% level in columns 6, 7, and 8.

32

that our conservative strategy allows us to only capture variation from firms that switch to managers with

different beliefs. The partial R2 is lower but of the same order of magnitude as other common predictors of

financial leverage, such as lagged one-year returns or tangibility (whose partial R2 is about 1%).

We also explore the inclusion of additional lags of stock returns. In unreported tests, we find that the

explanatory power of lagged stock returns declines as the time lag increases. The coefficient on Longholder CFO,

instead, remains very stable. In all cases, having a CFO Longholder in a firm predicts a significantly higher

market leverage ratio.40 The latter finding helps address concerns about insider information as an alternative

interpretation, i.e., the interpretation that Longholders are managers with positive inside information, who

may be reluctant to issue equity and choose high leverage. As discussed in Section 3, this concern is unlikely to

hold up since positive insider information should be transitory rather than persistent, and since we control for

the amount of vested options held at the same point in time. The inclusion of lagged returns (and Q) further

addresses this concern, as these controls are strong predictors of future returns. Nevertheless, the magnitude

or significance of the Longholder coefficient is unaffected.

In summary, our analysis of leverage confirms the empirical relevance of our findings regarding Prediction 1:

The influence of CFO overconfidence appears to be strong and persistent enough to translate into a measurable

influence even on the overall leverage ratio.

5 Overconfidence and the Cost of Debt

We now turn to our novel second prediction: CEO overconfidence is associated with a lower cost of debt, as

investors anticipate the continued commitment of an overconfident CEO to the investment, even in bad states of

the world. While it is not necessary for the empirical analysis that lenders recognize managerial overconfidence

and anticipate how it affects behavior, it is one possible interpretation. It is also implied by the evidence in

Otto (2014) on shareholders adjusting incentive contracts to the degree of overconfidence, and the evidence

40 The effect on market leverage is also significant in all specifications using Otto (2014)’s measure (see Appendix-Table C.3).A standard deviation increase in CFO overconfidence is associated with a 2.47% increase in leverage. Results are slightly weakerfor book leverage (perhaps because it is a noisier measure of the desired capital structure) but still positive in all specificationsand significant at the 5-10% level.

33

in Malmendier and Tate (2008) and Hirshleifer et al. (2012) on press portraits, suggesting that outsiders are

able to identify overconfident managers. For the empirical results, though, it suffices that investors predict the

managers’ effort choices correctly, even if attributed to a different reason, e.g., that they enjoy leisure less.

5.1 Empirical Strategy

We test this second prediction using DealScan data merged with our overconfidence measures. To match the

finer time periods in DealScan, we re-construct our firm-level control variables using the Compustat quarterly

database, following Valta (2012), among others. We measure the cost of debt financing as the spread between

the interest rate paid by the firm and the LIBOR (in basis points). This variable is slightly right-skewed, and

we employ the natural logarithm in our specifications. (Results are unaffected if we use the actual spread.)

We relate this outcome variable to managerial overconfidence as follows:

log(Net Interesti,t) = β1LTCEOi,t + β2LTCFOi,t +X ′i,tB + δt + εi,t, (8)

where LTCEOi,t and LTCFOi,t are our usual proxies for overconfidence (Longholder CEO and Longholder

CFO) and Xi,t is a vector of control variables at the manager, firm, and loan level, and also includes industry

fixed-effects. At the firm level, we include log(Assets) as larger firms might be perceived as less risky by

lenders;41 book leverage, given that highly indebted firms presumably face a higher cost of debt; cash holding

scaled by total assets as an additional proxy for a firm’s liquitidy; and the z-score, which captures the firm’s

default risk. Following Valta (2012), we also include earnings volatility, defined as the ratio of the standard

deviation of the past eight earnings changes to the average book assets over the past eight quarters. At the

loan level, we include log(Maturity) (in months) and log(Loan Amount) (in millions of dollars). We do not

have a prior on the signs of their coefficients. Loans with a longer horizon and for a higher amount may,

intuitively, be riskier, and so may be associated with higher spreads; however, in equilibrium, these may be

the loans made only to solid, safe firms. Finally, in some specifications we also add loan-type fixed effects. At

the managerial level, we include as usual the total number of shares and the number of vested options owned

41 We use log(Assets) for consistency with Valta (2012). Using our usual size proxy, log(Sales), produces the same results.

34

by each executive, standardized by the number of shares outstanding, to capture the moral hazard problem

generated by the separation of ownership and control. Finally, δt captures year-quarter fixed-effects.42

5.2 Baseline Results

Table 6 shows the main results of estimating equation (8). In this analysis, our prediction pertains to the role of

the CEO rather than the CFO since the actual implementation of an investment project, and its continuation

under adverse circumstances, rely first and foremost on the effort and decision-making of the CEO. Column

1 shows the baseline version of the estimation, which includes only Longholder CEO, industry fixed effects,

and year-quarter fixed effects as independent variables. We find that CEO overconfidence is associated with

a lower cost of debt. The coefficient of −0.199 is highly significant (p-value < 0.01). The estimated effect is

economically sizeable, amounting to about one fifth of a standard deviation of the outcome variable. Since

our dependent variable is log-transformed, we can interpret the coefficient as indicating a percentage change in

interest rates, i. e., a reduction of 19.1%, or 24.44 basis points relative to a sample mean of 127.97 basis points.

In column 2 we include the control variables mentioned above. Our coefficient of interest is slightly reduced

(−0.176), but the statistical significance increases, with a t-statistic over 3. Among the other regressors,

leverage and maturity enter with a positive sign, and size and loan amount are associated with lower interest

rates. Earnings volatility is associated with higher interest rates, albeit insignificantly. The same holds in

all other estimations shown in Table 6. (Only the coefficient of log(Maturity) becomes insignificant when we

include loan type dummies.) The managerial control variables for the CEO are insignificant and very small.

In columns 3 and 4 we turn to CFO overconfidence. We find some association between Longholder CFO

and lower interest rates in the baseline estimation, and the coefficient becomes marginally significant in the

specification with control variables. However, when we include our measures of CEO and CFO overconfidence

jointly (in columns 5 and 6), the association with CFO overconfidence becomes insignificant while the coefficient

42 Adam, Burg, Scheinert, and Streitz (Forthcoming) show that managerial optimism is related to the inclusion of performance-pricing provisions in loan contracts. Our results are unaffected when we include a dummy equal to one if any such provisions areincluded. Sunder, Sunder, and Tan (2010) find, instead, that loan contracts are more likely to include debt covenants when theborrower CEO is overconfident. Again, our results are very similar when we include a covenant dummy, identifying covenantsusing the procedure of Bradley and Roberts (2015).

35

of Longholder CEO is still large in magnitude and significant (−0.160, with a t-statistic of −2.989). Hence,

it appears that the effect of optimistic beliefs on banks’ willingness to finance a loan more cheaply does not

extend to the CFO. The interpretation offered by the model is that the CFO, while involved in financing

decisions, is less relevant to the operational decisions and effort choices pertaining to the implementation and

continuation of the project. In this sample, we fail to reject the hypothesis that the coefficients of CEO and

CFO overconfidence are equal at conventional significance levels. In the next subsection, we will see that the

difference is significant in the relevant subsample of firms with medium earnings variability, which is predicted

to drive the results.

We also note that the results persist even when we add loan-type fixed effects in column 7. The latter

specification is rather conservative, and has to be interpreted with some caution: A CEO’s beliefs may affect

financing costs also via the type of loan that financial intermediaries are willing to grant, as some types of

loans come with higher interest rates than others. Hence, the inclusion of controls for the type of loan may

absorb some of the relation between overconfidence and the cost of debt. Moreover, the analysis within loan

type is very demanding statistically, as our sample includes 18 different loan types.43 Nevertheless, we estimate

a similar effect. The coefficient on Longholder CEO is somewhat reduced (−0.116, corresponding roughly to a

reduction in interest rate spreads of about 15 basis points) but still significant at the 5% level.

Overall, overconfident CEOs appear to generate more favorable financing conditions. Longholder CFOs

affect the type of financing but not the cost of financing.

5.3 Effect of Overconfidence in Different Subsamples

A distinctive prediction of our models regards the specific type of firms that are able to obtain more favorable

debt financing under an overconfident CEO: firms that experience intermediate ranges of return variability. In

these firms, the uncertainty about future cash flows is large enough to reduce the incentives to ‘work hard’ on

the implementation or continuation of investment project in bad states of the world for rational CEOs, but

43 Indeed, the R2 of an estimation that excludes Longholder CEO is already very high in this specification (67.6%), which limitsthe additional variation that can be explained by the overconfidence proxy, which is 0.18%. The most common loan types are:revolving loans over more than one year (950 obs.), 364-days facilities (263 obs.), and generic term loans (124 obs.).

36

not for overoptimistic CEOs. Overconfidence drives a wedge into managerial choices as overconfident CEOs

continue to believe that they can generate a positive outcome when the intermediate signal is negative, and

rational CEOs do not. If, instead, uncertainty is very small or very large, there are no such differences in CEO

behavior. Rational and overconfident CEOs will either both continue or both abandon their investment efforts

upon negative intermediate news. Anticipating these choice, we do not predict differences in loan pricing.

To test the predicted non-monotonicity (in variability) of the effect of CEO overconfidence, we construct

several empirical proxies for firms’ return variability. A first natural proxy is earnings volatility, estimated

from actual earnings realizations. As discussed above, we use the ratio of the standard deviation of the past

eight earnings changes to average book assets over the past eight quarters, which is a popular proxy for profit

variability (at least) since Brealey, Hodges, and Capron (1976). (Recent uses include Valta (2012) and Matsa

(2010).) It is particularly suitable in our context, as it allows for earnings variability to vary over time, as the

managerial composition changes, and through a firm’s life cycle. In practice, we find that the correlation of

the volatility measure with its own lagged value at annual frequencies is about 78% in our data. That is, firms

appear to experience similar volatility over time, driven by the type of projects pursued in their industry or

firm. Hence, lagged values of volatility are strong predictors of future firm-level risk, making our measure of

return variability a good proxy for a firm’s risk from the lenders’ perspective.

It is also worth clarifying why we measure the volatility of earnings, not the volatility of individual project

returns for our empirical analysis. In our model, the firm’s investment consists of one project, and the two

types of volatility coincide. In practice, firms invest in more than one project, and the volatility of cash flows

from any single project is unlikely to fully determine the cost of financing. Instead, it is the occurrence of

firm-level shocks, as captured by the firm’s earnings, that induces or exacerbates the agent’s moral hazard

problem and lenders’ uncertainty about repayment. Hence, the volatility of overall earnings captures precisely

the mechanism the model illustrates: Lenders price the risk that, following a negative shock, a CEO will have

little incentive to carry through with a project.

In addition to employing earnings volatility, we use two measures that capture uncertainty as perceived by

outside observers: (1) analyst coverage, measured as the number of analysts who made at least one annual

earnings forecast and are included in IBES (similarly to Hong, Lim, and Stein (2000)); and (2) the coefficient

37

of variation of analysts’ annual earnings forecasts, defined as the standard deviation of forecasts normalized

by the absolute value of the mean forecast. As for the first, Schutte and Unlu (2009) show that analysts

coverage reduces noise, and Zhang (2006) finds that it is negatively related to price continuation anomalies,

suggesting that analyts reduce information uncertainty. Moreover, low analyst coverage is associated with

financial constraints (Whited and Wu (2006)), which in turn might indicate uncertainty regarding the firm’s

ability to repay their debt. As for the second measure, a large literature in accounting (cf. Cheng and Warfield

(2005)) shows that the coefficient of variation is associated with larger earnings surprises. One appealing

feature of this proxy for earnings variability is that it is not related to past earnings but to expectations of

future earnings, held by sophisticated market participants. For this last measure, we restrict our sample to firms

that are covered by at least ten analysts (896 observations). Both additional measures capture the uncertainty

a firm faces only indirectly as they rely on outsiders’ (analysts’) views, but provide useful robustness checks.

For each of our three proxies, we proceed as follows. First, we sort firms every year into a region of low,

medium, or high variability. We then estimate equation (8) on each of the three resulting subsamples, separately

for each of the three proxies. Our theoretical model does not pin down the thresholds between low, medium,

and high variability, and we use tercile splits as a natural starting point. Terciles allow us to test for the

predicted non-monotonicity while ensuring sufficient statistical power and estimates of comparable reliability

across subgroups.44 We check the robustness of our results to a wide range of different cutoffs. In addition to

the terciles, we show estimations using quantile splits of 35-30-35, 30-40-30, and 25-50-25, separately for each

measure, and we estimate a host of regressions where we further increase and decrease the top and bottom

ranges in 5%-steps. We also replicate the results using the continuous overconfidence proxy of Otto (2014).

The results are reported in Table 7. For brevity, we employ directly the empirical model with the full set of

controls, mirroring column 7 of Table 6, and report only the coefficients of Longholder CEO and Longholder

CFO. In all estimations, we continue to control for loan riskiness in multiple ways, as discussed above.

44 We divide the data into percentiles of each proxy year-by-year to account for non-stationary. (For example, analyst coveragehas increased steeply over time.) We then adjust the cutoffs to ensure that the size of each subsample is roughly one third, andsimilar for the other sample splits. The subsamples are otherwise not well balanced because the number of observations in someyears is small. The adjustment ensures that our results are not driven by variation in sample size as well as cutoff points, thoughthe results are similar without the adjustment. We also find similar results when we define quantiles using loans rather thancompanies, impose a minimum number of companies in a given year, or group together adjacent years with small sample sizes.

38

Starting from the earnings volatility proxy (Panel A), we see that the coefficient on Longholder CEO is

large and significant in the intermediate tercile, with a coefficient of -0.238 and a t-statistic of -3.057.45 In

terciles 1 and 3, instead, the coefficients on CEO overconfidence are small (-0.022 and -0.080) and insignificant.

In terms of economic magnitude, the estimate in the medium terciles implies that the spread a Longholder

CEO is charged is about 31 basis points lower than that of an unbiased manager. The difference between the

Longholder CEO coefficients in the low and medium sample is statistically significant at the 5% confidence

level, with a χ2-statistic equal to 4.958 computed under the null hypothesis of equality of the coefficients

(p-value= 0.026). We do not have enough power to reject the null hypothesis when we compare the CEO

Longholder coefficients in the medium and high sample (p-value= 0.105).

When using alternative sample splits, shown below the tercile split, we obtain qualitatively very similar

results, with the Longholder CEO coefficient being highly statistically significant only in the medium variability

subsample. We also replicate the result that the economic magnitude is always largest in the medium region.

We also obtain similar results when we use the two alternative proxies, analyst coverage and the coefficient

of variation (CV) of earnings estimates. In the case of analyst coverage, shown in Panel B, the estimated

effect of having an overconfident CEO on the cost of debt financing is large and significant, at the 1% level,

only in the medium range for the tercile split and all other quantile splits. In the low analyst-coverage and

the high analyst-coverage subsamples, instead, shown in columns 1 and 3 of Panel B, the Longholder CEO

coefficients are always small and insignificant. The same holds for the estimated effect of CFO overconfidence.

We note though that the differences in the estimated coefficients on CEO Longholder between subsamples are

significant at either 10% (low versus medium sample) or 1% level (medium versus high sample).

The CV-based estimates, shown in Panel C, also paint a similar picture. Because of the small sample size,

the estimates are slightly more imprecise. Here the CEO Longholder coefficient in the medium range is always

the largest in absolute value (except for the 30-40-30 split), although is significant only at the 10% level. We

note that the bottom range also features some negatively marginally significant estimates, even for the CFO.

The latter inconsistency reflects that the distribution of the coefficient of variation is very right-skewed in

our sample, with a median of 1.25% and a mean of 2.88%, so that the low and medium CV subsamples are

45 In this subsample, the increase in the R2 due to the inclusion of CEO overconfidence is 1.3% (74.5% versus 73.3%).

39

relatively similar in terms of the sorting variable.

Overall, these results, as well as the estimates from a host of regressions using alternative sample splits,46

reveal that CEO overconfidence predicts a willingness of banks to finance at lower costs only under medium

uncertainty, exactly as predicted by the model. The reliability of our results in this medium range, and the lack

thereof in the remaining subsamples, corroborates our theoretical interpretation. Furthermore, the difference

between these more precisely estimated CEO and CFO overconfidence coefficients is statistically significant:

Despite the larger sample size, the difference between the CEO and CFO coefficients was not significant in

Table 6, but is significant in almost half of the middle bins presented in Table 7 at at least the 10% level.47

One additional concern might be that our non-monotonicity result could reflect a more general difference in

statistical correlations between the cost of financing and the explanatory and control variables in the medium

range of variability. For example, interests rate might cluster at a lower bound and an upper bound in the other

ranges, but not in the medium range. However, the estimations show that, for example, the coefficient estimates

on Log(Firm Size) are statistically significant in the upper and lower bins at the 1% level in most regressions,

and at least at the 10% level in all regressions. Similarly, log(Loan Amount) is statistically significant at least

at the 5% level in the upper bin in all reported regressions, and in the lower bin in many cases. Moreover, in a

sequence of placebo tests in Appendix-Table C.10, we show that we do not replicate the results if we split the

sample by different sorting variables, which are not volatility-related: z-Score, leverage, and size (assets). None

of these sorting variables produces a sharp non-monotonic pattern. Instead, this pattern emerges as specific to

volatility, precisely as predicted by our model.48

The subsample results are central to our analysis in that they address concerns about unobserved covariates

more sharply. If an unobserved variable were to explain our findings, it would have to vary non-monotonically

with earnings volatility in order to rationalize our findings. In addition, any alternative interpretation of the

46 We replicate the results using the continuous overconfidence proxy of Otto (2014). As shown in Panels A and B of Appendix-Table C.5, the baseline CEO Longholder coefficients for the overall sample as well as the non-monotonicity and estimates in the“medium” sample under various sample splits are very similar. We do not observe similar patterns for CFO overconfidence andwhen using the other, more indirect, proxies for earnings volatility.

47 Of the twelve middle bin regressions, five differences are signifigant at the 10% level and two of these at the 5% level.48An additional result worth noting in Panel A of Appendix-Table C.10 is that the effect of CEO overconfidence appears

particularly strong in firms with high bankruptcy risk, as proxied by a low z-score. This could suggest that the CEO’s willingnessto work hard may reassure debtholders especially for firms in poor financial shape.

40

Longholder coefficients would also need to offer an explanation for why the CEO proxy and not the CFO

proxy is significant here, while the CFO proxy is significant for the financing choices analyzed above. This

variation is predicted by the model but hard to attribute to an unobserved variable being correlated with the

Longholder proxy. Taken together, the non-monotonicity and the variation in which Longholder proxy matters

seem unlikely to be explained by a hypothetical unobserved variable.

6 CFO Hiring Decisions

As the final step in our empirical analysis, we provide evidence on the third prediction, namely, that overconfi-

dent CEOs are more likely to hire similarly optimistic CFOs. We perform the analysis with the understanding

that the CEO does not select other top executives single-handedly, but is able to influence the board in the

selection of a CFO who will not systematically oppose her own vision (Landier et al. (2013)), and can strongly

affect the overall composition of the board (Shivdasani and Yermack (1999)).

As a first piece of suggestive evidence, we note that our measures of CEO and CFO overconfidence are

strongly correlated. The correlation coefficient is 25.3%, significant at the 1% level. However, as the CFO

may have been appointed before the CEO, the correlation might reflect firm effects or other factors outside

the CEO’s managerial choice. Thus, our main analysis focuses on CFOs appointed after a given CEO, and we

test whether a CFO is more or less likely to be overconfident depending on the CEO’s bias.

We identify all cases in which a firm in our data changes CFO, using ExecuComp’s execid identifier. We

then relate, for any new CFO appointed in year t, the time t Longholder CFO proxy to the time t−1 Longholder

CEO proxy and all relevant control variables at time t− 1. We estimate the following logit model:

Pr(LTCFOi,t = 1|LTCEOi,t−1, Xi,t−1, δt) = G(β LTCEOi,t−1 +X ′i,t−1B + δt + εi,t), (9)

where Xi,t−1 is a vector of controls, and δt a vector of year dummies. These filters leave us with 202 observations.

Results are reported in Table 8. In column 1, we include only CEO overconfidence and year fixed effects

as regressors. In column 2, we add industry fixed effects, which take into account the fact that overconfident

executives may sort into specific industries. For instance, Hirshleifer et al. (2012) find that overconfident CEOs

41

are more common in innovative industries.49 Column 3 adds our usual set of managerial controls, and column

4 also includes firm-level variables. Among the control variables, only CEO Vested Options are significant and

predict a lower probability of selecting an overconfident CFO. The inclusion of this variable does not diminish

the coefficient on Longholder CEO, though, but increases its size.

All four empirical models consistently show that overconfident CEOs are more likely to appoint overconfident

CFOs. Despite the small number of observations, the coefficient on Longholder CEO is always significant at

the 1% level. In our most demanding model (column 4), the estimates imply that an overconfident CEO is

over seven times more likely to hire an overconfident CFO relative to a rational CEO. Not surprisingly, given

the magnitude of this estimate, the incremental explanatory power of Longholder CEO is also large, with a

Pseudo R2 of 22.1%, relative to 15.1% when the overconfidence proxy is dropped.

Our results indicate that, above and beyond the direct influence of CEOs’ biased beliefs on corporate

outcomes, CEO biases exert an indirect influence via assortative matching. These results relate to the finding

of Landier et al. (2013) that firms who appoint a large fraction of executives to the board after the appointment

of the CEO tend to underperform their rivals. Our model is not designed, though, to explore the link between

firm value and (dis-)agreement among top managers, nor does the data provide measures of project quality.

Whether the relation between firm performance and board structure can be linked to CEO’s characteristics is

an interesting question for future research.

These results also help bridge the gap between the existing literature and our findings. In comparison

to Malmendier et al. (2011), we find a significantly smaller effect of CEO overconfidence on net debt issues

and leverage. To some extent, the differences might reflect different sample characteristics. In addition, the

assortative matching implies that some of the effect of CEO overconfidence on debt choice is a second-order

effect of hiring an overconfident CFO, which may not apply in all cases. That is, if overconfident CEOs hire

overconfident CFOs who then choose debt over equity, it is not surprising that we do not always find a strong

correlation between CEO overconfidence and capital structure choices. Although sample size plays a role, this

also helps to explain the lower significance of the CEO overconfidence findings in Malmendier et al. (2011) in

49 We include dummies for the Fama and French (1997) 12 industries classification rather than two-digit SIC Code industrydummies (as in the other tables) because of the small number of observations. However, using the more detailed industryclassification has no effect on our results.

42

comparison to our findings surrounding CFO overconfidence.

7 Conclusion

A key question in the analysis of managerial biases and the assessment of their empirical relevance is how to

account for the beliefs and choices of other managers with whom a biased manager interacts. Prior research

has mostly focused on one type of manager, typically the CEO, thereby running the risk of misattribution. For

example, the estimated impact of CEO overconfidence on financing choices might reflect, at least partly, the

influence of overconfident CFOs, who might assortatively match with overconfident CEOs.

In this paper, we have taken a first step towards advancing this line of research. We have considered the

beliefs of two key managers, the CEO and the CFO, jointly, in the context of financing decisions. We find

that CFOs’ behavioral traits have significant predictive power in explaining capital-structure decisions while

CEOs’ behavioral traits predict the cost of debt. Specifically, while firms with overconfident CFOs are more

likely to issue debt when accessing external capital, CFO biases are not predictive of the cost of financing.

Instead, the cost of debt financing depends on the type of CEO who runs the firm. Overconfident CEOs are

able to obtain cheaper debt financing than their rational peers. Finally, overconfident CEOs are more likely to

appoint overconfident CFOs. We provide a unifying theoretical framework that parsimoniously accommodates

these results.

Our findings corroborate previous research on managerial biases in corporate decisions, and point to the

importance of extending the analysis beyond the person of the CEO. We find that CEO overconfidence influ-

ences those corporate outcomes that are determined by CEOs, while CFO overconfidence does not. Similarly,

CFO overconfidence affects outcomes in the realm of the CFO and, here, CEO overconfidence does not matter.

This correlations helps address concerns about possible confounds of the Longholder proxy in prior research.

Our results also shed a new light on previous work on overconfidence that exclusively focused on CEOs.

In our (more recent) data, we estimate a weak influence of CEO traits on financing decisions when CEO’s

overconfidence is considered in isolation. We suggest that this is an indirect effect that occurs through CEO

hiring. The weaker estimates might also reflect that, as noted by practitioners, CFOs’ importance has increased

43

sharply over the past decade.50

Our findings point to at least two promising avenues for future research. First, it will be interesting to

explore the traits of other (C-suite) managers, such as CTOs or COOs, and their influence on corporate

decisions. Are their beliefs, biases, and personal characteristics associated with firm outcomes related to their

duties, and not associated with outcomes that do not fall into their decision-making realm? Such an analysis

will require a more comprehensive data with more detailed information on managers’ characteristics.

Second, our findings suggest richer economic implications of managerial characteristics than demonstrated

in previous research, including their influence on effort choices and on hiring decisions. Future research on

interaction and peer effects among managers that accounts for biased belief formation thus appears to be

another promising avenue. Relatedly, it might be interesting to explore how managerial traits and biases of

candidates affect how managers and boards make hiring decisions.

References

Acemoglu, D., P. Aghion, C. Lelarge, J. V. Reenen, and F. Zilibotti. 2007. Technology, Information, and the

Decentralization of the Firm. The Quarterly Journal of Economics 122:1759–1799.

Adam, T., V. Burg, T. Scheinert, and D. Streitz. Forthcoming. Managerial Optimism and Debt Contract

Design: The Case of Syndicated Loans. Management Science .

Aggarwal, R. K., and A. A. Samwick. 1999. The Other Side of the Trade-off: The Impact of Risk on Executive

Compensation. Journal of Political Economy 107:65–105.

Baker, M., and J. Wurgler. 2002. Market Timing and Capital Structure. Journal of Finance 57:1–32.

Baker, M., and J. Wurgler. 2013. Behavioral Corporate Finance: An Updated Survey, vol. 2 of Handbook of the

Economics of Finance, pp. 357–424. New York: Elsevier.

Bazerman, M., and D. A. Moore. 2013. Judgment in Managerial Decision Making. New York: Wiley.

Bazerman, M., and M. Neale. 1992. Nonrational Escalation of Commitment in Negotiation. European Man-

agement Journal 10:163–168.

50 For example, according to a 2014 survey of finance executives run by Accenture PLC, over 70% of the respondents believedthat “the CFO’s influence over executing business transformation initiatives has grown.” (See the report at https://www.

accenture.com/t20150523T035018Z__w__/us-en/_acnmedia/Accenture/Conversion-Assets/DotCom/Documents/Global/

PDF/Dualpub3/Accenture-2014-High-Performance-Finance-Study-CFO-Architect-Business-Value.PDF#zoom=50.)

44

Ben-David, I., and J. R. Graham. 2013. Managerial Miscalibration. The Quarterly Journal of Economics

128:1547–1584.

Ben-David, I., J. R. Graham, and C. R. Harvey. 2007. Managerial Overconfidence and Corporate Policies.

Working paper.

Benabou, R., and J. Tirole. 2002. Self-Confidence and Personal Motivation. The Quarterly Journal of Eco-

nomics 117:871–915.

Berk, J. B., and P. M. DeMarzo. 2007. Corporate Finance. New York: Pearson Education.

Bertrand, M., and A. Schoar. 2003. Managing with Style: The Effect of Managers on Firm Policies. The

Quarterly Journal of Economics 118:1169–1208.

Bester, C. A., and C. B. Hansen. 2016. Grouped Effects Estimators in Fixed Effects Models. Journal of

Econometrics 190:197–208.

Biais, B., and C. Casamatta. 1999. Optimal Leverage and Aggregate Investment. Journal of Finance 54:1291–

1323.

Bloom, N., R. Sadun, and J. Van Reenen. 2012. Americans Do IT Better: US Multinationals and the Produc-

tivity Miracle. The American Economic Review 102:167–201.

Boot, A. W., and A. V. Thakor. 2011. Managerial Autonomy, Allocation of Control Rights, and Optimal

Capital Structure. The Review of Financial Studies 24:3434–3485.

Bradley, M., and M. R. Roberts. 2015. The Structure and Pricing of Corporate Debt Covenants. The Quarterly

Journal of Finance 5:1–37.

Brealey, R. A., S. Hodges, and D. Capron. 1976. The Return on Alternative Sources of Finance. The Review

of Economics and Statistics 58:469–477.

Cai, J., J. L. Garner, and R. A. Walkling. 2009. Electing Directors. The Journal of Finance 64:2389–2421.

Camerer, C., and U. Malmendier. 2012. Behavioral Economics of Organizations. In P. Diamond and H. Var-

tiainen (eds.), Behavioral Economics and Its Applications. Princeton: Princeton University Press.

Campbell, T. C., M. Gallmeyer, S. A. Johnson, J. Rutherford, and B. W. Stanley. 2011. CEO Optimism and

Forced Turnover. Journal of Financial Economics 101:695–712.

Chava, S., and A. Purnanandam. 2010. CEOs versus CFOs: Incentives and Corporate Policies. Journal of

Financial Economics 97:263–278.

Chava, S., and M. R. Roberts. 2008. How Does Financing Impact Investment? The Role of Debt Covenants.

The Journal of Finance 63:2085–2121.

45

Cheng, Q., and T. D. Warfield. 2005. Equity Incentives and Earnings Management. The Accounting Review

80:441–476.

Chevalier, J., and G. Ellison. 1997. Risk Taking by Mutual Funds as a Response to Incentives. Journal of

Political Economy 105:1167–1200.

Datta, S., M. Iskandar-Datta, and K. Raman. 2001. Executive Compensation and Corporate Acquisition

Decisions. The Journal of Finance 56:2299–2336.

Dewatripont, M., and J. Tirole. 1994. A Theory of Debt and Equity: Diversity of Securities and Manager-

Shareholder Congruence. The Quarterly Journal of Economics 109:1027–1054.

Dittmar, A., and A. Thakor. 2007. Why do Firms Issue Equity? The Journal of Finance 62:1–54.

Englmaier, F. 2010. Managerial Optimism and Investment Choice. Managerial and Decision Economics

31:303–310.

Englmaier, F. 2011. Commitment in R&D Tournaments via Strategic Delegation to Overoptimistic Managers.

Managerial and Decision Economics 32:63–69.

Erickson, T., C. H. Jiang, and T. M. Whited. 2014. Minimum Distance Estimation of the Errors-in-Variables

Model using Linear Cumulant Equations. Journal of Econometrics 183:211–221.

Faccio, M., M.-T. Marchica, and R. Mura. 2016. CEO Gender, Corporate Risk-Taking, and the Efficiency of

Capital Allocation. Journal of Corporate Finance 39:193–209.

Fama, E. F., and K. R. French. 1997. Industry Costs of Equity. Journal of Financial Economics 43:153–193.

Fischer, P. E., J. D. Gramlich, B. P. Miller, and H. D. White. 2009. Investor Perceptions of Board Performance:

Evidence from Uncontested Director Elections. Journal of Accounting and Economics 48:172–189.

Galasso, A., and T. S. Simcoe. 2011. CEO Overconfidence and Innovation. Management Science 57:1469–1484.

Ge, W., D. Matsumoto, and J. L. Zhang. 2011. Do CFOs Have Style? An Empirical Investigation of the Effect

of Individual CFOs on Accounting Practices. Contemporary Accounting Research 28:1141–1179.

Gervais, S., and I. Goldstein. 2007. The Positive Effects of Biased Self-Perceptions in Firms. Review of Finance

11:453–496.

Gervais, S., J. B. Heaton, and T. Odean. 2011. Overconfidence, Compensation Contracts, and Capital Bud-

geting. The Journal of Finance 66:1735–1777.

Goel, A. M., and A. V. Thakor. 2008. Overconfidence, CEO Selection, and Corporate Governance. The Journal

of Finance 63:2737–2784.

46

Graham, J. R., and C. R. Harvey. 2001. The Theory and Practice of Corporate Finance: Evidence from the

Field. Journal of Financial Economics 60:187–243.

Graham, J. R., C. R. Harvey, and M. Puri. 2013. Managerial Attitudes and Corporate Actions. Journal of

Financial Economics 109:103–121.

Graham, J. R., C. R. Harvey, and M. Puri. 2015. Capital Allocation and Delegation of Decision-Making

Authority within Firms. Journal of Financial Economics 115:449–470.

Hackbarth, D. 2009. Determinants of Corporate Borrowing: A Behavioral Perspective. Journal of Corporate

Finance 15:389–411.

Hall, B. J., and K. J. Murphy. 2002. Stock Options for Undiversified Executives. Journal of Accounting and

Economics 33:3–42.

Hambrick, D. C., and P. A. Mason. 1984. Upper Echelons: The Organization as a Reflection of its Top

Managers. Academy of Management Review 9:193–206.

Hilary, G., C. Hsu, B. Segal, and R. Wang. 2016. The Bright Side of Managerial Over-Optimism. Journal of

Accounting and Economics 62:46–64.

Hirshleifer, D., A. Low, and S. H. Teoh. 2012. Are Overconfident CEOs Better Innovators? The Journal of

Finance 67:1457–1498.

Holmstrom, B., and J. Tirole. 1997. Financial Intermediation, Loanable Funds, and the Real Sector. The

Quarterly Journal of Economics 112:663–691.

Holmstrom, B., and J. Tirole. 1998. Private and Public Supply of Liquidity. Journal of Political Economy

106:1–40.

Hong, H., T. Lim, and J. C. Stein. 2000. Bad News Travels Slowly: Size, Analyst Coverage, and the Profitability

of Momentum Strategies. The Journal of Finance 55:265–295.

Huang, J., and D. J. Kisgen. 2013. Gender and Corporate Finance: Are Male Executives Overconfident Relative

to Female Executives? Journal of Financial Economics 108:822–839.

Jiang, J. X., K. R. Petroni, and I. Y. Wang. 2010. CFOs and CEOs: Who Have the Most Influence on Earnings

Management? Journal of Financial Economics 96:513–526.

Kaplan, S. N., M. M. Klebanov, and M. Sorensen. 2012. Which CEO Characteristics and Abilities Matter?

The Journal of Finance 67:973–1007.

Kim, J.-B., Y. Li, and L. Zhang. 2011. CFOs versus CEOs: Equity Incentives and Crashes. Journal of Financial

Economics 101:713–730.

Lakonishok, J., and I. Lee. 2001. Are Insider Trades Informative? Review of Financial Studies 14:79–111.

47

Landier, A., J. Sauvagnat, D. Sraer, and D. Thesmar. 2013. Bottom-up Corporate Governance. Review of

Finance 17:161–201.

Landier, A., and D. Thesmar. 2009. Financial Contracting with Optimistic Entrepreneurs. Review of Financial

Studies 22:117–150.

Malmendier, U. 2018. Behavioral Corporate Finance. In D. Bernheim, S. DellaVigna, and D. Laibson (eds.),

Handbook of Behavioral Economics, vol. 1, pp. 277–379. New York: Elsevier, North Holland.

Malmendier, U., and G. Tate. 2005. CEO Overconfidence and Corporate Investment. The Journal of Finance

60:2661–2700.

Malmendier, U., and G. Tate. 2008. Who Makes Acquisitions? CEO Overconfidence and the Market’s Reaction.

Journal of Financial Economics 89:20–43.

Malmendier, U., and G. Tate. 2015. Behavioral CEOs: The Role of Managerial Overconfidence. The Journal

of Economic Perspectives 29:37–60.

Malmendier, U., G. Tate, and J. Yan. 2011. Overconfidence and Early-Life Experiences: The Effect of Man-

agerial Traits on Corporate Financial Policies. The Journal of Finance 66:1687–1733.

Matsa, D. A. 2010. Capital Structure as a Strategic Variable: Evidence from Collective Bargaining. Journal

of Finance 65:1197–1232.

McDonald, R. L. 2005. Is It Optimal to Accelerate the Payment of Income Tax on Share-Based Compensation?

Working Paper.

Meikle, N. L., E. R. Tenney, and D. A. Moore. 2016. Overconfidence at Work: Does Overconfidence Survive

the Checks and Balances of Organizational life? Research in Organizational Behavior 36:121–134.

Otto, C. A. 2014. CEO Optimism and Incentive Compensation. Journal of Financial Economics 114:366–404.

Pagano, M., and P. F. Volpin. 2005. Managers, Workers, and Corporate Control. The Journal of Finance

60:841–868.

Pikulina, E., L. Renneboog, and P. Tobler. 2014. Overconfidence, Effort, and Investment. Working paper.

Puri, M., and D. T. Robinson. 2007. Optimism and Economic Choice. Journal of Financial Economics

86:71–99.

Ross, J., and B. M. Staw. 1993. Organizational Escalation and Exit: Lessons from the Shoreham Nuclear

Power Plant. Academy of Management Journal 36:701–732.

Schutte, M., and E. Unlu. 2009. Do Security Analysts Reduce Noise? Financial Analysts Journal 65:40–54.

Schwardmann, P., and J. van der Weele. 2017. Deception and Self-Deception. Working Paper.

48

Selody, K. 2010. Is the Risk Worth the Reward for Top Female Executives? Working Paper.

Sen, R., and R. Tumarkin. 2015. Stocking Up: Executive Optimism, Option Exercise, and Share Retention.

Journal of Financial Economics 118:399–430.

Shivdasani, A., and D. Yermack. 1999. CEO Involvement in the Selection of New Board Members: An Empirical

Analysis. The Journal of Finance 54:1829–1853.

Shroff, N. 2015. Real Effects of Financial Reporting Quality and Credibility: Evidence from the PCAOB

Regulatory Regime. Tech. rep.

Shyam-Sunder, L., and S. C. Myers. 1999. Testing Static Tradeoff against Pecking Order Models of Capital

Structure. Journal of Financial Economics 51:219–244.

Sunder, J., S. V. Sunder, and L. Tan. 2010. The Role of Managerial Overconfidence in the Design of Debt

Covenants. Working paper.

Tirole, J. 2006. The Theory of Corporate Finance. Princeton: Princeton University Press.

Tirole, J. 2010. The Theory of Corporate Finance. Princeton: Princeton University Press.

Valta, P. 2012. Competition and the Cost of Debt. Journal of Financial Economics 105:661–682.

Welch, I. 2004. Capital Structure and Stock Returns. Journal of Political Economy 112:106–131.

Whited, T. M., and G. Wu. 2006. Financial Constraints Risk. Review of Financial Studies 19:531–559.

Yim, S. 2013. The Acquisitiveness of Youth: CEO Age and Acquisition Behavior. Journal of Financial

Economics 108:250–273.

Zhang, X. F. 2006. Information Uncertainty and Stock Returns. The Journal of Finance 61:105–137.

49

8 Figures and Tables

Figure 1Timeline of the Model

t = −1CEO chooses

CFO.

t = 0CEO invests.CFO chooses

financing.

t = 1Signal about

projectprofitability.

t = 2CEO decideswhether toexert effort.

t = 3Cash flow is

realized.Initial investors

are repaid.

Table 1Data Construction

CEOs CFOs

Executives in ExecuComp 8,054 executives 7,402 executives

...matched with Thomson 3,372 executives 2,908 executives

Executives with at least10 transactions

1,623 executives 1,246 executives

...corresponding to... 13,898 firm-years 9,374 firm-years

Compustat sample with non-missingCEO/CFO Longholder

5,810 firm-years

Final sample (excluding financial,utilities and firms with missing controls)

4,581 firm-years

50

Table 2Summary Statistics

Panel A. Firm VariablesDebt Issues - Compustat (Table 3)

Variable Obs. Mean Median St. Dev.Net Debt Issue Indicator 2,939 0.507 1 0.500Q 2,939 2.396 1.816 2.121Profitability 2,939 0.178 0.172 0.150Tangibility 2,939 0.323 0.217 0.304log(Sales) 2,939 7.166 7.102 1.622Book Leverage 2,939 0.311 0.282 0.447

Financing Deficit and Leverage (Tables 4 and 5)Assets ($m) 4,581 5,792 1,643 14,465Sales ($m) 4,581 5,706 1,536 17,359Capitalization ($m) 4,581 8,311 2,264 20,864Net Financing Deficit ($m) 4,581 -254 -16 2,170Net Fin. Def. / Assets 4,581 -0.030 -0.018 0.366Net Debt Issues / Assets 4,581 0.027 0 0.159Book Leverage 4,557 0.289 0.257 0.432Q 4,581 2.416 1.874 1.960Change in Q 4,581 -0.034 0.030 1.628Profitability 4,581 0.185 0.174 0.140Change in Profitability 4,581 -0.002 0.002 0.097Tangibility 4,581 0.296 0.198 0.286Change in Tangibility 4,581 -0.007 -0.003 0.144log(Sales) 4,581 7.278 7.228 1.578Change in log(Sales) 4,581 0.108 0.097 0.221Market Leverage 4,557 14.570 10.559 15.364

Cost of Debt Financing (Tables 6 and 7)Interest Spread [bp] 1,651 127.970 100 102.497Loan Maturity [months] 1,651 46.409 60 21.778Loan Amount [$m] 1,651 590.82 300 1,080.37log(Assets) 1,651 7.951 7.841 1.377Book Leverage 1,651 0.234 0.23 0.15z-Score 1,651 3.585 2.452 4.475Earnings Volatility 1,651 0.018 0.008 0.072Cash Holding 1,651 0.122 0.062 0.191Analyst Coverage 1,651 12.009 10 7.600Coeff. Var. of Earn. Est. 896 0.029 0.013 0.064

Continued on next page

51

Table 2 – ContinuedPanel B. Manager Variables

Debt Issues - Compustat (Table 3)Variable Obs. Mean Median St. Dev.CEO Longholder 2,939 0.682 1 0.466CEO Stock Ownership [%] 2,939 1.882 0.341 0.467CEO Vested Options [%] 2,939 1.037 0.649 2.068CFO Longholder 2,939 0.529 1 0.499CFO Stock Ownership [%] 2,939 0.121 0.041 0.319CFO Vested Options [%] 2,939 0.260 0.129 0.772

Financing Deficit and Leverage (Tables 4 and 5)CEO Longholder 4,581 0.683 1 0.466CEO Stock Ownership [%] 4,581 1.806 0.305 4.839CEO Vested Options [%] 4,581 1.032 0.665 1.835CFO Longholder 4,581 0.530 1 0.499CFO Stock Ownership [%] 4,581 0.120 0.041 0.302CFO Vested Options [%] 4,581 0.249 0.128 0.644

Cost of Debt Financing (Tables 6 and 7)CEO Longholder 1,651 0.665 1 0.472CEO Stock Ownership [%] 1,651 1.318 0.283 3.876CEO Vested Options [%] 1,651 0.869 0.587 1.097CFO Longholder 1,651 0.543 1 0.498CFO Stock Ownership [%] 1,651 0.114 0.040 0.349CFO Vested Options [%] 1,651 0.214 0.113 0.489

52

Table 3Debt Issues (Compustat)

Table 3 shows the estimated log odds ratios from logistic regressions. The binary dependent variable isequal to 1 if Net Debt Issues during the year are positive. Net Debt Issues is long-term debt issuanceminus long-term debt reduction. Longholder CEO/Longholder CFO is a binary variable where 1 signifiesthat the CEO/CFO at some point during his tenure held exercisable options until the last year beforeexpiration, given that the options were at least 40% in the money entering their last year. We requiremanagers to have at least ten transactions recorded in Thomson Reuters to be included in the sample.Stock Ownership is option-excluded shares held by the CEO/CFO as a percentage of common sharesoutstanding. Vested Options is the number of exercisable options held by the CEO/CFO as a percentageof common shares outstanding. Q is the book value of assets plus the market value of equity minus thebook value of equity minus deferred tax, divided by the book value of assets. Profitability is operatingincome before depreciation divided by lagged assets. Tangibility is property, plants and equipmentdivided by lagged assets. Book Leverage is the sum of current liabilities and long-term debt dividedby the sum of current liabilities, long-term debt and book equity. Stock Ownership, Vested Options,Q, Profitability, Tangibility, log(Sales), and Book Leverage are measured at the beginning of the year.2-digit SIC level industry fixed-effects are included in all regressions. Standard errors are clusteredby firm, and corresponding t-statistics are shown in parentheses. ***, **, and * indicate statisticallydifferent from zero at the 1%, 5%, and 10% level of significance, respectively.

(1) (2) (3) (4) (5) (6) (7)Longholder CEO 0.111 -0.007 0.012 0.020 -0.126

(0.944) (-0.058) (0.096) (0.170) (-1.086)Longholder CFO 0.354*** 0.392*** 0.352*** 0.412*** 0.437***

(3.182) (3.430) (3.062) (3.510) (3.725)CEO Shares -0.008 -0.028* -0.009

(-0.568) (-1.735) (-0.659)CEO Vested -0.005 -0.007 0.037

Options (-0.144) (-0.084) (1.131)Q -0.058 -0.059 -0.060

(-1.486) (-1.395) (-1.423)Profitability 0.731 0.706 0.706

(1.213) (1.184) (1.179)Tangibility 0.274 0.296 0.324

(0.949) (1.021) (1.103)log(Sales) 0.478*** 0.475*** 0.477***

(9.757) (9.873) (9.856)Book Leverage 0.096 0.100 0.093

(0.687) (0.709) (0.677)CFO Shares -0.085 -0.182 -0.078

(-0.621) (-1.055) (-0.582)CFO Vested -0.120 -0.577** -0.193**

Options (-1.451) (-2.486) (-2.237)Manager Ctrl. NO YES NO YES NO YES YESFirm Controls NO YES NO YES NO NO YESIndustry FE YES YES YES YES YES YES YESYear FE NO YES NO YES NO YES YESObservations 2,939 2,939 2,939 2,939 2,939 2,939 2,939Pseudo R2 0.042 0.163 0.047 0.169 0.047 0.107 0.170

53

Table 4Financing Deficit

Table 4 presents the estimates of OLS regressions with Net Debt Issues normalized by assets at the beginningof the year as the dependent variable. Net Debt Issues is long-term debt issuance minus long-term debtreduction. Longholder CEO (LTCEO) and Longholder CFO (LTCFO) are binary variables where 1 signifiesthat the CEO or CFO, at some point during their tenure, held exercisable options until the last year beforeexpiration, given that the options were at least 40% in the money entering their last year. FD is the NetFinancing Deficit, which is defined as cash dividends plus investment plus change in working capital minuscash flow after interest and taxes, normalized by assets at the beginning of the year, which is identical tothat in Malmendier, Tate and Yan (2011). Manager-level control variables include Stock Ownership andVested Options. Stock Ownership is option-excluded shares held by the CEO or CFO as a percentage ofcommon shares outstanding. Vested Options is the number of exercisable options held by the CEO orCFO as a percentage of common shares outstanding. Firm-level control variables include changes in Q,Profitability, Tangibility and log(Sales). Q is the book value of assets plus the market value of equity minusthe book value of equity minus deferred tax, divided by the book value of assets. Profitability is operatingincome before depreciation divided by assets at the beginning of the year. Tangibility is property, plantsand equipment divided by assets at the beginning of the year. Manager-level and firm-level control variablesare all measured at the beginning of the year. Columns (3), (6), and (9) also include the interaction of NetFinancing Deficit with the control variables including year dummies, so the standalone variable is redundantand its coefficient omitted from the table. Standard errors are clustered by firm, and corresponding t-statistics are shown in parentheses. ***, **, and * indicate statistically different from zero at the 1%, 5%,and 10% level of significance, respectively.

(1) (2) (3) (4) (5) (6) (7) (8) (9)FD × LTCEO 0.024 0.054 0.164** -0.024 0.011 0.104*

(0.207) (0.545) (2.112) (-0.244) (0.129) (1.781)FD × LTCFO 0.243** 0.210* 0.179*** 0.247** 0.208** 0.166***

(2.151) (2.026) (2.981) (2.269) (2.110) (2.958)FD 0.203** 0.158** 0.106*** 0.099*** 0.118* 0.094*

(2.317) (2.498) (2.828) (2.803) (1.837) (1.716)LTCEO -0.003 -0.001 0.003 -0.005 -0.001 0.003

(-0.295) (-0.083) (0.274) (-0.488) (-0.141) (0.313)LTCFO 0.012 0.010 -0.003 0.013 0.008 -0.008

(0.850) (0.704) (-0.259) (0.908) (-0.568) (-0.616)

Manager Contr. NO YES YES NO YES YES NO YES YESFirm Controls NO YES YES NO YES YES NO YES YESFD x Controls NO NO YES NO NO YES NO NO YESFirm FE YES YES YES YES YES YES YES YES YESYear FE NO YES YES NO YES YES NO YES YESObservations 4,581 4,581 4,581 4,581 4,581 4,581 4,581 4,581 4,581R-squared 0.208 0.294 0.447 0.272 0.337 0.482 0.273 0.338 0.490

54

Table 5Leverage

Table 5 presents the estimates of OLS regressions with market leverage (multiplied by 100) as dependentvariable. Market leverage is long-term debt plus debt in current liabilities item, all divided by price timescommon shares outstanding plus the numerator. Longholder CEO (LTCEO) and Longholder CFO (LTCFO)are binary variables where 1 signifies that the CEO or CFO, at some point during their tenure, held exercisableoptions until the last year before expiration, given that the options were at least 40% in the money enteringtheir last year. Stock Ownership is option-excluded shares held by the CEO or CFO as a percentage ofcommon shares outstanding. Vested Options is the number of exercisable options held by the CEO or CFO asa percentage of common shares outstanding. Firm-level control variables include Q, Profitability, Tangibility,log(Sales) and Net Financing Deficit. Q is the book value of assets plus the market value of equity minusthe book value of equity minus deferred tax, divided by the book value of assets. Profitability is operatingincome before depreciation divided by lagged assets. Tangibility is property, plants and equipment divided bylagged assets. Manager-level and firm-level control variables are all measured at the beginning of the year.Net Financing Deficit (FD) which is cash dividends plus investment plus change in working capital minuscash flow after interest and taxes, normalized by lagged assets. Returnst−1 are lagged one year returns. Allthe regressions include year and firm fixed-effects. Standard errors are clustered by firm, and correspondingt-statistics are shown in parentheses. ***, ** and * indicate statistically different from zero at the 1%, 5%,and 10% level of significance, respectively.

(1) (2) (3) (4) (5) (6) (7) (8)LTCEO 1.485 0.890 1.063 0.483 0.444 0.382

(1.127) (0.690) (0.794) (0.371) (0.344) (0.298)LTCFO 4.151*** 3.678** 4.044*** 3.800*** 3.681*** 3.730***

(3.045) (2.815) (2.972) (2.904) (2.831) (2.874)CEO Shares 0.062 0.107* 0.102* 0.107*

(0.908) (1.798) (1.686) (1.776)CEO Vested 0.140 0.125 0.115 0.114

Options (1.376) (1.150) (1.055) (1.020)CFO Shares -0.482 -0.795 -0.758 -0.697

(-1.132) (-1.438) (-1.396) (-1.318)CFO Vested -0.330* 0.208 0.213 0.208

Options (1.800) (1.096) (1.132) (1.127)Q -0.676*** -0.658*** -0.654*** -0.758*** -0.633***

(-4.260) (-4.209) (-4.205) (-4.422) (-3.859)Profitability -15.100*** -15.245*** -145.206*** -14.660*** -14.111***

(-5.542) (-5.627) (-5.601) (-5.361) (-5.053)Tangibility 6.825*** 6.894*** 6.901*** 6.759*** 6.737***

(4.688) (4.757) (4.769) (4.531) (4.446)log(Sales) 3.049*** 3.009*** 3.066*** 3.282*** 3.096***

(4.117) (4.085) (4.143) (4.333) (4.087)FD 2.906*** 2.967***

(4.253) (4.343)Returnst−1 -0.918***

(-4.446)

Manager Controls NO YES NO YES NO YES YES YESFirm Controls NO YES NO YES NO YES YES YESFirm FE YES YES YES YES YES YES YES YESYear FE YES YES YES YES YES YES YES YESObservations 4,557 4,557 4,557 4,557 4,557 4,557 4,557 4,557R2 0.089 0.142 0.094 0.147 0.095 0.148 0.161 0.169

55

Table 6Cost of Debt Financing

Table 6 presents regressions of log(Interest Spread) on our overconfidence measures and several controlvariables, including year-quarter and industry fixed-effects. Log(Interest Spread) is the difference be-tween the interest rate of the loan in basis points and the London Interbank Offered Rate. LongholderCEO (LTCEO) and Longholder CFO (LTCFO) are binary variables where 1 signifies that the CEOor CFO, at some point during their tenure, held exercisable options until the last year before expira-tion, given that the options were at least 40% in the money entering their last year. Stock Ownershipis option-excluded shares held by the CEO or CFO as a percentage of common shares outstanding.Vested Options is the number of exercisable options held by the CEO or CFO as a percentage of com-mon shares outstanding. Book Leverage is (long-term debt + debt in current liabilities) / (long-term+ debt in current liabilities + common equity). Z-Score is 1.2 × (current assets - current liabilities)+ 1.4 × retained earnings + 3.3 × pretax income + 0.6 × (market capitalization / total liabilities) ×total assets + 0.9 × sales, all scaled by total assets. Cash holding is cash and short-term investmentsdivided by total assets. Earnings Volatility is the standard deviation of the past eight earnings changesto the average book asset size over the past eight quarters. Control variables are all measured at thebeginning of the year. Standard errors are clustered by firm, and corresponding t-statistics are shownin parentheses. ***, **, and * indicate statistically different from zero at the 1%, 5%, and 10% level ofsignificance, respectively.

(1) (2) (3) (4) (5) (6) (7)Longholder CEO -0.199*** -0.176*** -0.185** -0.160*** -0.116**

(-2.817) (-3.586) (-2.477) (-2.989) (-2.386)Longholder CFO -0.096 -0.098* -0.040 -0.048 -0.040

(-1.288) (-1.899) (-0.520) (-0.884) (-0.835)Log(Size) -0.214*** -0.218*** -0.214*** -0.208***

(-7.812) (-7.785) (-7.801) (-8.130)Leverage 0.621*** 0.615*** 0.617*** 0.553***

(6.051) (6.013) (6.033) (5.982)Z-Score -0.011 -0.013* -0.011 -0.010

(-1.571) (-1.719) (-1.616) (-1.467)Log(Amount) -0.098*** -0.096*** -0.097*** -0.091***

(-3.631) (-3.624) (-3.660) (-3.529)Earnings Volatility 0.335 0.360 0.332 0.423

(1.120) (1.119) (1.119) (1.567)Cash Holding 0.153 0.116 0.150 0.106

(1.265) (1.018) (1.245) (0.991)CEO Shares 0.002 0.002 0.004

(0.476) (0.452) (0.679)CEO Vested Options 0.019 0.005 0.005

(1.102) (0.236) (0.230)CFO Shares 0.023 0.029 0.041

(0.359) (0.433) (0.682)CFO Vested Options 0.057 0.051 0.046

(1.284) (1.035) (1.111)Manag. Controls NO YES NO YES NO YES YESFirm Controls NO YES NO YES NO YES YESIndustry FE YES YES YES YES YES YES YESYear-Quarter FE YES YES YES YES YES YES YESLoan Type FE NO NO NO NO NO NO YESObservations 1,651 1,651 1,651 1,651 1,651 1,651 1,651R-Squared 0.417 0.632 0.410 0.627 0.417 0.633 0.682

56

Table 7Net Interest Rates Across Subsamples (Different Cutoffs)

Panels A, B and C test the relation between CEO overconfidence and the cost of debt acrossdifferent subsamples, using different cutoffs for low, medium, and high variability in eachsorting variable (Earnings Volatility in Panel A, Analysts Coverage in Panel B, and Coefficientof Variation of Earnings Forecasts in Panel C). All panels show regressions of log(Interest RateSpread) on our measures of overconfidence and the same control variables and fixed effects asin Column 7 of Table 6. We estimate the empirical model specified in equation 8 in the maintext in each subsample. ***, ** and * indicate statistically different from zero at the 1%, 5%,and 10% level of significance, respectively.

Panel A. Sorting by Earnings Volatility(1) (2) (3)

Bottom Tercile Medium Tercile Top TercileLongholder CEO -0.022 -0.238*** -0.080

(-0.268) (-3.057) (-0.844)Longholder CFO -0.128* -0.013 0.055

(-1.746) (-0.179) (0.666)Observations 550 547 554R-Squared 0.779 0.741 0.778

Bottom 35% Medium 30% Top 35%Longholder CEO -0.040 -0.181** -0.079

(-0.506) (-2.105) (-0.866)Longholder CFO -0.095 -0.052 0.051

(-1.371) (-0.645) (0.656)Observations 574 500 577R-Squared 0.778 0.753 0.770

Bottom 30% Medium 40% Top 30%Longholder CEO 0.003 -0.226*** -0.169*

(0.033) (-3.146) (-1.825)Longholder CFO -0.142* -0.013 0.096

(-1.753) (-0.199) (1.145)Observations 492 662 497R-Squared 0.781 0.733 0.790

Bottom 25% Medium 50% Top 25%Longholder CEO -0.014 -0.214*** -0.153

(-0.159) (-3.061) (-1.605)Longholder CFO -0.110 -0.021 0.093

(-1.317) (-0.328) (1.082)Observations 417 822 412R-Squared 0.799 0.718 0.833

Continued on next page

57

Table 7 – ContinuedPanel B. Sorting by Analysts’ Coverage

(1) (2) (3)Bottom Tercile Medium Tercile Top Tercile

Longholder CEO -0.065 -0.242*** 0.070(-0.899) (-3.117) (0.776)

Longholder CFO 0.032 -0.082 -0.117*(0.442) (-1.027) (-1.668)

Observations 559 554 538R-Squared 0.705 0.745 0.791

Bottom 35% Medium 30% Top 35%Longholder CEO -0.092 -0.245*** 0.075

(-1.288) (-3.128) (0.879)Longholder CFO 0.023 -0.104 -0.113

(0.332) (-1.272) (-1.621)Observations 591 481 579R-Squared 0.703 0.758 0.790

Bottom 30% Medium 40% Top 30%Longholder CEO -0.079 -0.230*** 0.082

(-1.083) (-3.210) (0.936)Longholder CFO 0.036 0.006 -0.097

(0.452) (0.075) (-1.370)Observations 491 666 494R-Squared 0.731 0.706 0.802

Bottom 25% Medium 50% Top 25%Longholder CEO -0.067 -0.212*** 0.085

(-0.783) (-3.200) (0.954)Longholder CFO -0.005 0.031 -0.142*

(-0.058) (0.461) (-1.782)Observations 413 809 429R-Squared 0.764 0.697 0.812

Continued on next page

58

Table 7 – ContinuedPanel C. Sorting by Coefficient of Variation of Earnings Estimates

(1) (2) (3)Bottom Tercile Medium Tercile Top Tercile

Longholder CEO -0.200* -0.243* 0.164(-1.965) (-1.950) (1.056)

Longholder CFO -0.188* -0.095 -0.252*(-1.881) (-0.900) (-1.690)

Observations 294 300 297R-Squared 0.865 0.851 0.794

Bottom 35% Medium 30% Top 35%Longholder CEO -0.188* -0.251* 0.154

(-1.979) (-1.729) (0.994)Longholder CFO -0.186* -0.120 -0.235

(-1.850) (-0.955) (-1.549)Observations 309 273 310R-Squared 0.859 0.865 0.787

Bottom 30% Medium 40% Top 30%Longholder CEO -0.205 -0.147 0.201

(-1.576) (-1.403) (1.222)Longholder CFO -0.159 -0.067 -0.178

(-1.507) (-0.768) (-1.086)Observations 261 364 267R-Squared 0.889 0.823 0.795

Bottom 25% Medium 50% Top 25%Longholder CEO -0.095 -0.161* 0.087

(-0.593) (-1.843) (0.401)Longholder CFO -0.146 -0.060 -0.097

(-1.188) (-0.737) (-0.518)Observations 223 450 219R-Squared 0.897 0.789 0.834

59

Table 8CFO Hiring

Table 8 shows the estimated log odds ratios from logit regressions with Longholder CFO as thedependent variable. The sample includes all instances in which a new CFO is appointed betweenyear and year and the following variables are not missing: (i) the overconfidence proxy for the newCFO at time; (ii) the overconfidence proxy for the incumbent CEO at time; (iii) manager and firmcontrol variables at time. Longholder CEO/Longholder CFO is a binary variable where 1 signifiesthat the CEO/CFO at some point during his tenure held exercisable options until the last yearbefore expiration, given that the options were at least 40% in the money entering their last year.We require managers to have at least ten transactions recorded in Thomson Reuters to be includedin the sample. Stock Ownership is option-excluded shares held by the CEO as a percentage ofcommon shares outstanding. Vested Options is the number of exercisable options held by the CEOas a percentage of common shares outstanding. Q is the book value of assets plus the market valueof equity minus the book value of equity minus deferred tax, divided by the book value of assets.Profitability is operating income before depreciation divided by assets at the beginning of the year.Tangibility is property, plants and equipment divided by assets at the beginning of the year. BookLeverage is the sum of current liabilities and long-term debt divided by the sum of current liabilities,long-term debt and book equity. Stock Ownership, Vested Options, QQ, Profitability, Tangibility,Log(Sales), and Book Leverage are measured at the beginning of the year. Year fixed effects areincluded in all the regressions and Fama and French (1997) 12 industry dummies effects are includedin columns 2-4. All standard errors are adjusted for clustering at the firm level. ***, **, and *indicate statistically different from zero at the 1%, 5%, and 10% level of significance, respectively.

(1) (2) (3) (4)Longholder CEO 1.124*** 1.436*** 2.031*** 2.010***

(2.791) (3.247) (4.413) (4.309)CEO Vested Options -0.780*** -0.792***

(-3.404) (-2.980)CEO Shares -0.027 -0.024

(-0.765) (-0.658)Q -0.071

(-0.454)Profitability 1.705

(0.890)Tangibility 1.484*

(1.764)Log(Sale) -0.023

(-0.161)Book Leverage 0.064

(0.177)

Manager Controls NO NO YES YESFirm Controls NO NO NO YESIndustry FE NO YES YES YESYear FE YES YES YES YESObservations 202 202 202 202Pseudo R-squared 0.085 0.143 0.205 0.221

60

Appendix

This Appendix consists of three parts. Appendix A provides the proofs referenced in Section 2 of the paper.Appendix B lists detailed definitions of the variables in our empirical analysis. Appendix C provides summarystatistics for specific subsamples of the data, as well as numerous robustness checks.

A Proofs

We prove Propositions 1, 2, and 3 in subsections A.1, A.3, and A.4, respectively. In subsection A.2, we definethe optimal equity contract, which is a necessary step to prove Propositions 2 and 3. In subsection A.5, wediscuss the robustness of our theoretical results to different parametric assumptions.

A.1 Optimal Debt Contract

Proof of Proposition 1. We show that the face value of debt offered to overconfident and rational CEOsis identical for the case of low variability, with D∗ω = I; that it is lower for overconfident CEOs than rationalCEOs for the case of intermediate variability, with D∗ω = I and D∗0 = I + σ, respectively; and that it is againidentical for the case of high variability, with D∗ω = I + σ. The ranges of return variability σ are σ ≤ ∆ − Bfor low return variability; ∆−B + ω ≥ σ > ∆−B for intermediate return variability; and σ > ∆−B + ω forhigh return variability.

First, we show jointly that D∗ω = I for the case of low variability (σ ≤ ∆− B) and also, when the CEO isoverconfident, for the case of intermediate variability (∆− B + ω ≥ σ > ∆− B). We can summarize the fullrange of both cases as σ ≤ ∆−B + ωCEO.

We start by showing that the CEO’s IC constraint (4b) is satisfied in both states of the world. In the goodstate, the CEO exerts effort iff

max {0, I + σ + ∆ + ωCEO − I} ≥ max {0, I + σ − I}+B

⇐⇒ max {0, σ + ∆ + ωCEO} ≥ max {0, σ}+B

⇐⇒ σ + ∆ + ωCEO ≥ σ +B

⇐⇒ ∆ + ωCEO ≥ B,

(A.1)

which holds given our model assumption ∆ > B. In the bad state, the CEO exerts effort iff

max {0, I − σ + ∆ + ωCEO − I} ≥ max {0, I − σ − I}+B

⇐⇒ max {0,−σ + ∆ + ωCEO} ≥ max {0,−σ}+B

⇐⇒ −σ + ∆ + ωCEO ≥ B

⇐⇒ ∆−B + ωCEO ≥ σ,

(A.2)

which is exactly the parameter range we are considering. Thus, the CEO exerts effort in both states.We can now plug these effort choices into the participation constraint (4c), and obtain

1

2(min {I, I + σ + ∆}+ min {I, I − σ + ∆}) = I, (A.3)

61

i.e., the participation constraint holds with equality since σ ≤ ∆ − B + ωCEO and, per model assumption,B ≥ ω, and thus σ < ∆. Hence, under D∗ω = I, all surplus goes to existing shareholders, which in turn impliesthat the (perceived) firm value is maximized under this contract. The expected utility of a rational CFO is ∆,whereas the overconfident CFO expects to get (∆ + ω).

To prove uniqueness, consider any other contract with face value D ≷ I. We can rule out D < I, as it doesnot satisfy the participation constraint. For D > I, there are two cases to consider: either the CEO exertseffort in both states of the world, or she does not. If she does, the surplus is the same as under D∗ω = I anddebtholders extract positive rents. Hence this type of contract cannot be optimal for the CFO. If she doesnot, the resulting welfare loss implies that the rents that the CFO can extract (under debtholders’ break-evenconstraint) will not be maximized. Hence, D∗ω = I is optimal when σ ≤ ∆−B + ωCEO.

Second, we show that D∗ω = I + σ for the case of high variability (σ > ∆− B + ω) and, when the CEO isrational, also for the case of intermediate variability (∆−B + ω ≥ σ > ∆−B). We can summarize these twocases as σ > ∆−B + ωCEO.

We start again from the IC constraint (4b) and show that, under D∗ω = I + σ, the CEO exerts effort in thegood state and shirks in the bad state. In the good state, the CEO exerts effort iff

max {0, I + σ + ∆ + ωCEO − I − σ} ≥ max {0, I + σ − I − σ}+B

⇐⇒ max {0,∆ + ωCEO} ≥ max {0, 0}+B

⇐⇒ ∆ + ωCEO ≥ B,

(A.4)

which is implied by our initial assumption ∆ > B. In the bad state, the CEO shirks iff

max {0, I − σ + ∆ + ωCEO − I − σ} < max {0, I − σ − I − σ}+B

⇐⇒ max {0,−2σ + ∆ + ωCEO} < max {0,−2σ}+B

⇐⇒ max {0,−2σ + ∆ + ωCEO} < B.

(A.5)

This is satisfied both if −2σ+∆+ ωCEO ≤ 0 since 0 < B; and if −2σ+∆+ ωCEO > 0 since, over the parameterrange σ > ∆ − B + ωCEO, it must also hold that 2σ > ∆ − B + ωCEO, and hence −2σ + ∆ + ωCEO < B.Therefore, under D∗ω = I + σ, the CEO exerts effort in the good state of the world and shirks in the bad stateof the world.

Turning to the participation constraint (4c) and plugging in these effort choices, we can now show that theparticipation constraint holds with equality:

1

2(min {I + σ, I + σ + ∆}+ min {I + σ, I − σ}) = I. (A.6)

Again, debtholders receive I in expectation, and all surplus goes to existing shareholders. In this case, arational CFO’s expected utility is ∆/2, and an overconfident CFO expects to get (∆ + ω)/2.

To see that this is an optimal contract, and that it is the unique optimal contract, consider an alternativecontract D 6= D∗ω. We can again rule out D < I since debtholders would not break even. For D ≥ I, we firstask in which state of the world the CEO would exert effort under such a contract. In the bad state of theworld, the CEO exerts effort under contract D iff

max{

0, I − σ + ∆ + ωCEO − D}≥ max

{0, I − σ − D

}+B. (A.7)

62

With D ≥ I, the IC becomes max{

0, I − σ + ∆ + ωCEO − D}≥ B, (A.8)

which holds only if I − D ≥ σ − (∆ + ωCEO − B). However, as we are analyzing the parameter space ofσ− (∆ + ωCEO −B) > 0, this implies I − D > 0, contradicting that D ≥ I. Hence, the CEO shirks in the badstate of the world, and we are left with two cases: Either the CEO exerts effort only in the good state of theworld, or in neither state. Because debtholders cannot obtain more than I − σ in the bad state of the world,the participation constraint requires D ≥ D∗ω = I + σ in order for debtholders to break even. As D 6= D∗ω, wemust have D > D∗ω. Thus, if the CEO exerts effort only in the good state of the world, debtholders extract astrictly positive rent (given the higher face value D > D∗ω), contradicting optimality. And if the CEO exertseffort in neither state, the contract with face value D∗ω generates higher total surplus for the CFO because ofthe CEO’s higher effort choice (in the good state of the world), in combination with the lower face value. Thiscontradicts optimality. �

A.2 Optimal Equity Contract and Cost of Equity

As an intermediate step in the analysis of the CFO’s choice between debt and equity, we first define in Lemma1 the optimal equity contract, conditional on equity financing, and discuss the resulting cost of equity. As inthe case of debt, we will see that the optimal equity contract is independent of the CFO’s type.

We adopt the same notation as for the debt contract. Let πCFO(S, e) be the return to the project underthe CFO’s beliefs. We denote the fraction of the firm owned by new shareholders as γ. The CFO solves thefollowing program to determine the (second-best) optimal equity contract:

maxγ

(1− γ)E[πCFO(S, eS)] (A.9a)

uCEO(S, γ, eS) ≥ uCEO(S, γ, e′S)∀S and eS 6= e′S (A.9b)

γE[π(S, eS)] ≥ I (A.9c)

Lemma 1 (Optimal Equity Contract). The optimal equity contract depends on the CEO’s but not on theCFO’s bias. In particular, we have

γ∗ω = II+∆

and eS = 1∀S if ∆+ωCEO

I+∆∆ ≥ B and

γ∗ω = 1 and eS = 0∀S if ∆+ωCEO

I+∆∆ < B.

Proof of Lemma 1. We start from the IC constraint under equity financing, shown in inequality (3) in thepaper. We know from (3) that the CEO’s choice of effort is independent of the state of the world. She exertseffort in both states iff (1− γ)(∆ + ωCEO) ≥ B ⇐⇒ γ ≤ 1− B

∆ + ωCEO(A.10)

In this case, the participation constraint of new shareholders becomes

γ(I + ∆) ≥ I (A.11)

Conversely, she does not exert effort in either state of the world if and only if γ > 1 − B∆+ωCEO

. In thelatter case the participation constraint becomes γ ≥ 1, and the only feasible equity financing contract assignsfull ownership to new shareholders, while the CFO obtains zero payoff. In the former case, instead, theparticipation constraint is satisfied with equality, γ∗ω = I

I+∆, and the resulting (perceived) payoff of the CFO

is (1− γ∗ω)E[πCFO(S, 1)] = ∆I+∆

(I + ∆ + ωCFO) = (∆ + ∆I+∆

ωCFO) > 0.

63

Hence, inducing effort is optimal if γ∗ω = II+∆

satisfies the IC constraint, i.e., if II+∆≤ 1− B

∆+ωor, solving

for B, if B ≤ ∆+ωCEO

I+∆∆. If, instead, ∆+ωCEO

I+∆∆ < B, the CEO cannot be induced to exert effort under any

equity contract that allows new shareholders to break even. Therefore, the project is going to deliver I inexpectation and the only contract satisfying equity holders’ participation constraint requires γ∗ω = 1. �

A.3 Choice between Debt and Equity

We show that an overconfident CFO is weakly more likely to issue debt than a rational CFO, whether the CEOis overconfident or rational. Specifically, there are parameter ranges for which an overconfident CFO strictlyprefers debt while a rational CFO does not (and is instead indifferent between the two financing choices).51

Whenever the overconfident CFO strictly prefers equity, instead, so does the rational CFO.The proof of Proposition 2 involves comparing the CFO’s perceived utility under debt and equity financing.

We use again the notation ωCEO to capture both the case of a rational CEO (ωCEO = 0) and of an overcon-fident CEO (ωCEO = ω). As before, “perceived firm value” is short-hand for “expected payoff to incumbentshareholders under the CFO’s beliefs.”

Proof of Proposition 2. Recall from the proof in Appendix A.2 that the optimal equity contract dependson whether ∆+ωCEO

I+∆< B or not. This holds whether the CEO is rational or overconfident.

If ∆+ωCEO

I+∆∆ < B, the optimal equity contract assigns all surplus to new shareholders (γ∗ = 1), and the

CEO shirks in both states of the world. We have also shown that the optimal debt contract induces the CEOto exert effort in at least in one state of the world, achieving a strictly higher firm value, and that not allsurplus goes to the lenders. Since investors must break even (under any type of financing), the gain in firmvalue translates into rents to incumbent shareholders, and thus to the CFO. Therefore, both types of CFOsprefer debt financing over the parameter range ∆+ωCEO

I+∆∆ < B.

If instead ∆+ωCEO

I+∆∆ ≥ B, the optimal equity contract does not assign all surplus to new shareholders, and

the CEO exerts effort in both states of the world. As a result, a rational and an overconfident CFO havedifferent perceptions of the value created by the CEO:

i. Rational CFO. Under the optimal equity contract, incumbent shareholders obtain (1− I/[I + ∆])(I +∆) = ∆. Under the optimal debt contract, we have to consider two cases: If σ ≤ ∆ − B + ωCEO, the CEOexerts effort in both states of the world, and the expected firm value is (I + σ + ∆ + I − σ + ∆)/2− I = ∆. Ifσ > ∆− B + ωCEO, the CEO exerts effort only in the good state of the world, and the expected firm value is(I + σ + ∆ + ∆− I − σ)/2 = ∆/2. Comparison of these firm values gives us the CFO’s choice, shown in thefirst table (“Rational CFO”) below.

ii. Overconfident CFO. The overconfident CFO believes incorrectly that the CEO’s effort is worth∆ + ω instead of ∆. Thus, as the CEO exerts effort in both states of the world under equity financing, theCFO perceives firm value to incumbent shareholders under equity financing to be (1 − I

I+∆)(I + ∆ + ω) =

∆ + ∆I+∆

ω. The same misperception applies under debt financing when σ ≤ ∆ − B + ωCEO: As the CEOexerts effort in both states of the world, and the face value of debt is I, the CFO perceives firm value to equal(I + σ + ∆ + ω + I − σ + ∆ + ω)/2 − I = ∆ + ω. If instead σ > ∆ − B + ωCEO, the CEO shirks in the badstate of the world, and the CFO’s perceived firm value is therefore (I + σ + ∆ + ω − I − σ)/2 = (∆ + ω)/2.The next table below (“Overconfident CFO”) summarizes these computations and the CFO’s choices.

51 If the rational CFO randomizes his financing choice when indifferent, with positive probabilities for both debt and equity, anoverconfident CEO uses strictly more debt, on average, than a rational CFO over this parameter range.

64

Rational CFO Debt EquityPreferred

Choice

∆+ωCEO

I+∆∆ ≥ B andσ ≤ ∆−B + ωCEO ∆ ∆ Indifferent

∆+ωCEO

I+∆∆ ≥ B andσ > ∆−B/σ + ωCEO ∆/2 ∆ Equity

Overconfident CFO Debt EquityPreferred

Choice

∆+ωCEO

I+∆∆ ≥ B andσ ≤ ∆−B + ωCEO ∆ + ω ∆ + ∆

I+∆ω Debt

∆+ωCEO

I+∆∆ ≥ B andσ > ∆−B + ωCEO (∆ + ω)/2 ∆ + ∆

I+∆ω Equity

In summary, for either rational or overconfident CEOs, we find that both types of CFOs choose debtfinancing for some parameter ranges (∆+ωCEO

I+∆∆ < B), and both types choose equity financing for other ranges

(∆+ωCEO

I+∆∆ ≥ B andσ > ∆− B + ωCEO). However, we also find that in some instances only the overconfident

CFO strictly prefers debt (∆+ωCEO

I+∆∆ ≥ B andσ ≤ ∆−B + ωCEO). In other words:

- If the rational CFO strictly prefers debt, so does the overconfident CFO.- If the rational CFO is indifferent between debt and equity, the overconfident CFO strictly prefers debt.- If the rational CFO strictly prefers equity, so does the overconfident CFO.Taken together, these results imply that, conditioning on the CEO’s type, an overconfident CFO weakly prefersdebt relative to a rational CFO. �

A.4 Hiring Decision

Proof of Proposition 3. The CEO is indifferent between the two types of CFOs if she expects either typeto make the same financing choice. Therefore, we only need to analyze cases in which the two types of CFOsmay behave differently, given the CEO’s bias.

We start by considering the rational CEO’s choice (ωCEO = 0). From Section A.3 above we know that if∆2

I+∆∆ ≥ B and σ ≤ ∆−B, the overconfident CFO strictly prefers debt (see A.3.ii) but the rational CFO does

not (see A.3.i); he is indifferent. The rational CEO, instead, is always indifferent between a debt and an equitycontract, as she expects to obtain ∆ under either contract. Therefore, she will not exhibit any preferenceregarding the CFO to be appointed.

Moving to an overconfident CEO’s choice (ωCEO = ω), from Section A.3 above, we know that if ∆+ωI+∆

∆ ≥ Band σ ≥ ∆−B + ω, the rational CFO is indifferent between debt and equity, whereas the overconfident CFOstrictly prefers debt. With debt financing, the overconfident CEO expects to obtain (∆ + ω); with equity her

65

perceived future payoff is only (∆+ ∆I+∆

ω). Therefore, under the CEO’s beliefs, debt strictly dominates equity,and she prefers an overconfident CFO, who chooses debt financing for sure, to a rational CFO, who instead,being indifferent, may choose equity.

In sum, a rational CEO is indifferent between appointing an overconfident or a rational CFO; an overcon-fident CEO weakly prefers an overconfident CFO.52 �

A.5 Robustness of the Theoretical Results

We now provide a detailed discussion of the robustness of our results to reverting either of our two assumptionsregarding the extent of the moral hazard problem for the rational CEO (∆ > B) and for the overconfidentCEO (B ≥ ω).

a. Assume B ≥ ∆. If B > ∆, a rational CEO never exerts effort. The optimal debt contract is thusD∗0 = I + σ. Similarly, the optimal debt contract is γ∗0 = 1, and the CEO will not exert effort either. Inboth cases, the value of the project to incumbent shareholders is zero. In the knife-edge case B = ∆, it is stillpossible to induce the rational CEO to exert high effort in the good state of the world, but only under a debtcontract, by keeping her indifferent between shirking and working hard (again D∗0 = I + σ).

The overconfident CEO, instead, can still be induced to exert effort if B > ∆, namely, as long as ω ≥ B−∆.Under the optimal contract, she will work hard, either in both states of the world or only in the good one, atleast under a debt contract.

Hence, by altering the assumption ∆ > B, we affect the rational CEO’s effort decision, but not the maininsight that overconfidence can ameliorate conditional financing terms.

b. Assume ω > B. In our main analysis, the assumption ω ≤ B implies that the discrepancy in beliefsbetween the overconfident CEO and debtholders is not “too large” and that, whenever the CEO exerts effort,she does not default. We analyze how removing this assumption affects the optimal debt contract and CFO’schoice between debt and equity.

b.i) Optimal debt contract. For ω > B, there is an additional case to consider under debt financing: Theoverconfident CEO may exert effort in the bad state of the world. Consider the incentive constraint

max {0, I − σ + ∆ + ωCEO −D} ≥ max {0, I − σ −D}+B (A.12)

There are two subcases. First, suppose that ∆ < σ ≤ ∆− 1/2(B − ω). In this case, the optimal contract forthe overconfident CEO requires D∗ω = I + σ −∆. Plugging D∗ω into the constraint (A.12) we get

max {0, I − σ + ∆ + ω − (I + σ −∆)} ≥ max {0, I − σ − (I + σ −∆)}+B (A.13)

or (2∆− 2σ + ω) ≥ B, (A.14)

which is satisfied under ∆ < σ ≤ ∆ − 1/2(B − ω). Hence, the overconfident CEO mistakenly expects not todefault after exerting effort, but debtholders correctly anticipate that they will receive only I − σ + ∆ in thebad state of the world. At the same time, the IC A.1 is satisfied, delivering I + σ −∆ to debtholders in thegood state of the world. Therefore, debtholders will break even in expectation. The proofs of optimality and

52 As in Appendix-Section A.3, we use the expression ‘weakly prefers’ because we have not specified how to break indifferencebetween debt and equity. If we assume that a CFO randomizes between the two financing choices whenever indifferent, with somepositive weight on the debt choice, an overconfident CEO will strictly prefer an overconfident CFO to a rational one.

66

uniqueness are similar to those in subsection A.1 and are omitted for brevity.Now consider the subcase σ > ∆− 1/2(B−ω). Here, it is not possible to induce the overconfident CEO to

exert effort and simultaneously ensure debtholders to break even. Intuitively, any debt contract that induceseffort in the bad state of the world would require a face value of debt that is too low to satisfy debtholders’participation constraint.

Without making any assumption on the relative size of ω and B, we conclude that the optimal debt contractfor an overconfident CEO is given by:

- D∗ω = I + σ ifσ > ∆−B + ω or ∆− 1/2(B − ω) < σ ∧ σ > ∆;- D∗ω = I + σ −∆ if ∆− 1/2(B − ω) ≥ σ > ∆;- D∗ω = I if ∆−B + ω ≥ σ ∧∆ ≥ σ.

Thus, although the optimal debt contract becomes more complicated in this more general case, the basic insightof Proposition 1 remains unaffected, with overconfidence reducing the cost of debt when profit variability islarge but not extreme.b.ii) Financing choice. Moving to the analysis of the CFO’s choice between debt and equity, we find thatif ω > B, the different structure of the optimal debt contract can affect the overconfident CFO’s preferencebetween debt and equity whenever:

(i) the CEO is overconfident, with bias ω;(ii) ∆+ω

I+∆∆ ≥ B (i.e., equity financing is available with γ∗ω = I/(I + ∆)); and

(iii) ∆− 1/2(B/− ω) ≥ σ > ∆.In this case, the rational CFO is indifferent between debt and equity. The reason is that he correctly anticipatesthat the CEO defaults in the bad state of the world but, because of the lower cost of debt, firm value will stillbe maximized. In particular, the unbiased expected value of the firm is (I + σ + ∆ + 0− (I + σ −∆))/2 = ∆.This is equivalent to the firm value obtained under an equity contract, making him indifferent between the twofunding choices. For an overconfident CFO (who shares the bias ω of the CEO) the perceived expected firmvalue under optimal debt contract D∗ω = I+σ−∆ equals (I+σ+∆+I−σ+∆+ω)/2−(I+σ−∆) = 2∆+ω−σ.Therefore, he (weakly) prefers debt if

2∆ + ω − σ ≥ ∆ +∆

I + ∆ω, (A.15)

Without further assumptions we cannot establish whether A.15 holds or not. Notice, however, that thisinequality reduces to

ωI

I + ∆≥ σ −∆ (A.16)

The left-hand side of this expression is increasing in ω. This means that we can always find a sufficiently largevalue for ω such that A.16 holds. In particular, we can exploit the fact that σ ≤ I. Replacing σ = I in A.16and rearranging terms, we get

ω ≥ I − ∆2

I(A.17)

In other words, the overconfident CFO displays a preference for debt for sufficiently high overconfidence,with expression A.17 providing a lower bound for ω. Note that this kind of indeterminacy result for certainparameter ranges is common when debt is very risky (see for example the model in Malmendier et al. (2011)).Here, however, the main contribution is to distinguish the role of CEO and CFO’s traits, with the latter

67

dominating in financing choices.

68

B Variables Definitions

Below, we provide detailed definitions of the variables used in the empirical analyses. For the variables extractedfrom Compustat, ExecuComp and Dealscan we also indicate the data item (in italic).

Table B.1Variables Definitions and Sources

Variable DefinitionManager Variables (constructed from Thomson Insider Filing Dataset, CRSP and ExecuComp)LTCEO/LTCFO binary variable where 1 signifies that the CEO/CFO at some point during his

tenure held exercisable options until the last year before expiration, giventhat the options were at least 40% in the money entering their last year

Stock Ownership option-excluded shares (shrown excl opts) held by the CEO/CFO as apercentage of common shares outstanding (csho)

Vested Options number of exercisable options (opt unex exer num) held by the CEO/CFO asa percentage of common shares outstanding (csho)

Firm Variables (constructed from Compustat (Annual or Quarterly), SDC, Dealscan)Net Debt Issues ($m) long term debt issuance (dltis) - long term debt reduction (dltr)Net Debt Issues Indicator(Compustat)

binary variable where 1 signifies that Net Debt Issues during the year ispositive

Net Debt Issues Indicator(SDC)

binary variable where 1 signifies that the company issued bonds during theyear

Book Leverage (long-term debt (dltt) + debt in current liabilities (dlc)) / (long-term debt(dltt) + debt in current liabilities (dlc) + common equity (ceq))

Net Financing Deficit($m) cash dividends (dv) + investment + change in working capital - cash flowafter interest and taxes, where . . .

...investment capx + ivch + aqc + fuseo - sppe - siv for firms with cash flow format code(scf ) 1 to 3;capx + ivch + acq - sppe - siv - ivstch - ivaco for firms with cash flow formatcode 7;0 for other firms

...change in working capital wcapc + chech + dlcch for firms with cash flow format code 1;-wcapc + chech - dlcch for firms with cash flow format code 2 and 3;-recch - invch - apalch - txach - aoloch + chech - fiao - dlcch for firms withcash flow format code 7;0 for other firms

...cash flow after interestand taxes

ibc + xidoc + dpc + txdc + esubc + sppiv + fopo + fsrco for firms with cashflow format code 1 to 3;items ibc + xidoc + dpc + txdc + esubc + sppiv + fopo + exre for firms withcash flow format code 7;0 for other firms

Book Leverage (long-term debt (dltt) + debt in current liabilities (dlc)) / (long-term debt(dltt) + debt in current liabilities (dlc) + common equity (ceq))

Continued on next page

69

Table B.1 – ContinuedVariable DefinitionMarket Leverage (long-term debt (dltt) + debt in current liabilities (dlc)) / (price (prcc) ×

common shares outstanding (csho) + debt in current liabilities (dlc) +long-term debt (dltt))

Q (assets (at) + price (prcc) × common shares outstanding (csho) - commonequity (ceq) - balance sheet deferred taxes and investment tax credit (txditc))/ assets (at)

Profitability operating profit (oibdp) / lagged assets (at)Changes in Profitability profitability - lagged profitabilityTangibility property, plants and equipment (ppent) / lagged assets (at)Changes in Tangibility tangibility - lagged tangibilitylog(Sales) log(sales (sale))Changes in log(Sales) log(sales) - lagged log(sales)log(Interest Spread) difference between the interest rate the borrower pays in basis points and the

London Interbank Offered Rate (variable allindrawn in Dealscan)Z-Score 1.2 × (current assets (actq) - current liabilities (dlcq)) / total assets (atq) +

1.4 × (retained earnings (req) / total assets (atq)) + 3.3 × (pretax income(piq) / total assets (atq)) + 0.6 × (market capitalization (cshoq × prccq) /total liabilities (ltq)) + 0.9 × (sales (saleq) / total assets (atq))

Earnings Volatility (standard deviation of the past eight earnings (ibq) changes) / (average bookasset size over the past eight quarters).

log(Amount) log (natural logarithm) of the amount of the loan (in million dollars) (amt)Analysts’ Coverage number of analysts making at least one annual earnings forecast in a given

yearCoefficient of Variation ofEarnings Estimates

standard deviation of annual earnings forecasts normalized by the absolutevalue of the mean forecast (We require at least ten forecasts made.)

70

C Robustness Checks

Appendix C presents additional details about the data and a series of robustness checks for all estimationspresented the paper.

In Appendix-Table C.1 we show the descriptive statistics for our largest sample (employed in Tables 4and 5 in the main text) split by the four possible combinations of executives’ biases, as identified by theLongholder measure: (a) both executives are classified as overconfident; (b) the CEO is rational and the CFOis overconfident; (c) the CEO is overconfident and the CFO rational CEO; and (d) both are overconfident.

Appendix-Tables C.2−C.6 show the estimation results if we use Otto (2014)’s continuous empirical measureof CEO overconfidence. Under this approach, overconfidence is measured as the weighted average of transaction-specific overconfidence dummmies. We first classify each option exercise of an executive. The transaction-specific dummy takes the value 1 if the options were exercised within one year of their expiration date andwere at least 40% in the money at the end of the preceding year. Otherwise, the dummy takes the value 0. Wethen average the value of the optimism dummies for each executive across his or her transactions, weightingeach observation by the number of options that were exercised. Therefore, the final overconfidence measuretakes values between 0 and 1. We repeat all of our empirical analyses using this measure and show the resultsbelow, omitting the coefficients on the control variables for brevity. The specifications and the control variablesare exactly the same, except in Appendix-Table C.6 (CFO Hiring) where, given the nature of our dependentvariable, we estimate a Tobit rather than a logit model. The continuous Longholder proxies are normalized bytheir respective sample standard deviations for ease of interpretation.

Appendix-Table C.7 reports our estimations of the relation between managerial overconfidence and propen-sity to issue debt using SDC data rather than Compustat. To generate the data for this analysis, we matchall issues of debt, equity, and hybrid securities (convertible debt and convertible preferred stock) with theExecuComp-Compustat merged sample described in the main text. We have 694 observations when no con-trol variables are required, and 647 observations in the subsample where all control variables are available.Moreover, as the industry dummies perfectly predict some of the debt issues, the actual sample usable foridentification varies further, between 694 and 585 observations (when the full set of controls is included). Theestimations mirror those using the Compustat data in the main text (in Table 3): We estimate again a logitmodel with a dummy equal to one if a firm issued debt in a given year, and 0 otherwise, i. e., if the firm issuedequity or hybrid securities, and using the same control variables. Given the small sample, we choose to displaythe estimations using all available observations for the respective specifications. Results of this exercise confirmthose in the main text.

As mentioned in the main text, we have also re-run all our tests by restricting the analysis to firms thathave appreciated by more than 40% in the previous ten years. This robustness check has the limitation thatit mechanically excludes from the sample all firms that, in any given year, have been listed for less than 10years. We show the replications of Tables 3 and 6 in the main text (Tables C.8 and C.9 of this Appendix).

As a robustness test for the results regarding the cost of debt, we replicate our tests of Table 7 using, assorting variables, z-score, leverage and size. In Table C.10, we find that none of these cross-sectional cuts areable to reproduce the non-monotonic pattern of the Longholder coefficient shown in Table 7, where we sort byproxies for volatility.

Finally, we check the robustness of our results to varying the minimum number of transactions. In our maintests, we require CEOs and CFOs to have at least 10 transactions recorded. Figure C.1 plots the coefficientsof interest in each regression using an array of minimum transaction requirements between 1 and 10. We onlyplot only the coefficients from the most conservative regressions (last column of each table).

71

Table C.1Summary Statistics Split by Executives’ Bias

Panel A. Panel B.Both Executives Overconfident Rational CEO, Overconfident CFO

Variable Obs. Mean Median St. Dev. Obs. Mean Median St. Dev.

Assets ($m) 1,928 6,009 1,710 14,230 499 5,739 1,467 11,608Sales ($m) 1,928 7,018 1,575 23,654 499 4,622 1,203 8,395Capitalization ($m) 1,928 8,822 2,268 24,762 499 7,576 2,006 14,265Net Fin. Deficit ($m) 1,928 -258 -14 1,739 499 50 -23 2,886Net Fin. Deficit / Assets 1,928 -0.010 -0.016 0.305 499 -0.041 -0.023 0.418Net Debt Issues / Assets 1,928 0.031 0.000 0.169 499 0.028 0.000 0.191Book Leverage 1,922 0.304 0.278 0.395 498 0.255 0.254 0.213Q 1,928 2.264 1.785 1.735 499 2.449 1.972 2.029Change in Q 1,928 -0.022 0.023 1.389 499 -0.111 0.016 1.698Profitability 1,928 0.185 0.177 0.124 499 0.191 0.180 0.132Change in Profitability 1,928 -0.003 0.002 0.078 499 -0.005 0.002 0.112Tangibility 1,928 0.321 0.201 0.302 499 0.278 0.201 0.305Change in Tangibility 1,928 -0.007 -0.003 0.115 499 -0.005 -0.003 0.285log(Sales) 1,928 7.355 7.263 1.589 499 7.227 7.023 1.543Change in log(Sales) 1,928 0.097 0.093 0.221 499 0.091 0.074 0.226Market Leverage 1,922 0.154 0.122 0.152 498 0.131 0.104 0.131CEO Stock Ownership (%) 1,928 2.070 0.374 4.809 499 1.147 0.186 4.937CEO Vested Options (%) 1,928 1.023 0.696 1.254 499 0.690 0.396 0.906CFO Stock Ownership (%) 1,928 0.149 0.056 0.282 499 0.106 0.043 0.178CFO Vested Options (%) 1,928 0.303 0.165 0.877 499 0.234 0.143 0.274

Continued on next page

72

Table C.1 – ContinuedPanel C. Panel D.

Overconfident CEO, Rational CFO Both Executives RationalVariable Obs. Mean Median St. Dev. Obs. Mean Median St. Dev.

Assets ($m) 1,199 6,578 1,951 17,990 955 4,394 1,307 10,766Sales ($m) 1,199 5,250 1,758 11,508 955 4,197 1,328 10,327Capitalization ($m) 1,199 9,803 2,879 21,782 955 5,788 1,767 11,801Net Fin. Deficit ($m) 1,199 -470 -13 2,828 955 -134 -21 1,410Net Fin. Deficit / Assets 1,199 -0.049 -0.015 0.374 955 -0.041 -0.024 0.432Net Debt Issues / Assets 1,199 0.022 0.000 0.122 955 0.026 0.000 0.159Book Leverage 1,186 0.256 0.213 0.307 954 0.317 0.279 0.661Q 1,199 2.621 1.999 2.233 955 2.449 1.819 1.960Change in Q 1,199 -0.011 0.036 2.004 955 -0.049 0.054 1.507Profitability 1,199 0.191 0.175 0.156 955 0.176 0.161 0.153Change in Profitability 1,199 0.000 0.000 0.115 955 0.000 0.005 0.100Tangibility 1,199 0.283 0.198 0.260 955 0.274 0.192 0.268Change in Tangibility 1,199 -0.008 -0.004 0.098 955 -0.006 -0.002 0.137log(Sales) 1,199 7.362 7.377 1.531 955 7.043 7.073 1.613Change in log(Sales) 1,199 0.120 0.104 0.212 955 0.124 0.108 0.229Market Leverage 1,186 0.127 0.079 0.143 954 0.159 0.109 0.176CEO Stock Ownership (%) 1,199 1.775 0.316 4.676 955 1.652 0.253 5.014CEO Vested Options (%) 1,199 1.188 0.758 2.718 955 1.032 0.598 1.790CFO Stock Ownership (%) 1,199 0.080 0.023 0.270 955 0.120 0.039 0.408CFO Vested Options (%) 1,199 0.199 0.085 0.481 955 0.212 0.110 0.307

73

Table C.2Debt Issues (Compustat)

Logit regressions with the Net Debt Issues Indicator as the dependent variable, regressed on Otto (2014)’soverconfidence measure (normalized by its sample standard deviation) for CEOs and CFOs and the samecontrol variables as in Table 3. ***, **, and * indicate statistically different from zero at the 1%, 5%, and10% level of significance, respectively.

(1) (2) (3) (4) (5) (6) (7)Longholder CEO 0.088 0.095 0.071 0.050 0.050

(1.565) (1.407) (1.222) (0.812) (0.714)Longholder CFO 0.076 0.140*** 0.052 0.081 0.133**

(1.363) (2.617) (0.883) (1.400) (2.253)Manager Controls NO YES NO YES NO YES YESFirm Controls NO YES NO YES NO NO YESIndustry FE YES YES YES YES YES YES YESYear FE NO YES NO YES NO YES YESObservations 2,938 2,938 2,938 2,938 2,938 2,938 2,938Pseudo R-Squared 0.044 0.153 0.047 0.157 0.047 0.099 0.157

74

Table C.3Financing Deficit

Replication of the estimation of Table 4 with Otto (2014)’s overconfidence measure (normalized by its samplestandard deviation) for CEOs and CFOs and the same control variables as in Table 4. ***, **, and * indicatestatistically different from zero at the 1%, 5%, and 10% level of significance, respectively.

(1) (2) (3) (4) (5) (6) (7) (8) (9)FD × Longh. CEO 0.005 0.011 0.046 -0.024 -0.007 -0.015

(0.099) (0.226) (1.107) (-0.374) (-0.114) (-0.511)FD × Longh. CFO 0.045 0.017 0.027 0.064 0.022 0.034

(0.497) (0.221) (0.762) (0.614) (0.239) (0.862)FD 0.207*** 0.175*** 0.194*** 0.180*** 0.209*** 0.184***

(3.277) (3.525) (3.165) (3.058) (3.453) (3.455)Longholder CEO -0.008 -0.005 0.001 -0.008 -0.005 0.002

(-1.234) (-0.690) (0.127) (-1.467) (-0.735) (0.560)Longholder CFO -0.003 -0.004 -0.003 -0.001 -0.003 -0.004

(-0.831) (-1.006) (-0.663) (-0.214) (-0.663) (-0.872)Manager Controls NO YES YES NO YES YES NO YES YESFirm Controls NO YES YES NO YES YES NO YES YESFD × Controls NO NO YES NO NO YES NO NO YESFirm FE YES YES YES YES YES YES YES YES YESYear FE NO YES YES NO YES YES NO YES YESObservations 4,581 4,581 4,581 4,581 4,581 4,581 4,581 4,581 4,581R2 0.208 0.291 0.438 0.229 0.303 0.496 0.233 0.303 0.499

Table C.4Leverage

OLS regressions with market leverage as dependent variable, regressed on Otto (2014)’s overconfidence mea-sure (normalized by its sample standard deviation) for CEOs and CFOs and the same control variables as inTable 5. ***, **, and * indicate statistically different from zero at the 1%, 5%, and 10% level of significance,respectively.

(1) (2) (3) (4) (5) (6) (7) (8)Longholder CEO 1.265* 0.592 0.838 0.185 0.192 0.179

(1.863) (0.921) (1.244) (0.293) (0.290) (0.266)Longholder CFO 2.499*** 2.223*** 2.305*** 2.203*** 2.175*** 2.185***

(5.248) (4.718) (4.448) (4.256) (4.183) (4.166)Manager Contr. NO YES NO YES NO YES YES YESFirm Controls NO YES NO YES NO YES YES YESReturnt−1 NO NO NO NO NO NO NO YESFirm FE YES YES YES YES YES YES YES YESYear FE YES YES YES YES YES YES YES YESObservations 4.552 4.552 4.552 4.552 4.552 4.552 4.552 4.552R-squared 0.090 0.143 0.093 0.145 0.094 0.146 0.158 0.166

75

Table C.5Cost of Debt Financing

Panel A shows regressions of log(Interest Spread) on Otto (2014)’s overconfidence measures (normalized byits sample standard deviation) for CEOs and CFOs and the same control variables and fixed effects as inTable 6. Log(Interest Spread) is the difference (in basis points) between the interest rate the borrower paysand the LIBOR. In Panel B we test the relation between CEO overconfidence and the cost of debt acrossdifferent subsamples, using different cutoffs for low, medium, and high Earnings Volatility. The controlsvariables are as in column (7) of Panel A. ***, ** and * indicate statistically different from zero at the 1%,5%, and 10% level of significance, respectively. .

Panel A. Baseline Regressions

(1) (2) (3) (4) (5) (6) (7)Longholder CEO -0.050 -0.064*** -0.063* -0.063*** -0.053**

(-1.478) (-3.044) (-1.791) (-2.758) (-2.475)Longholder CFO 0.023 -0.019 0.042 -0.000 -0.003

(0.779) (-1.060) (1.364) (-0.014) (-0.158)Manag. Controls NO YES NO YES NO YES YESFirm Controls NO YES NO YES NO YES YESIndustry FE YES YES YES YES YES YES YESYear-Quarter FE YES YES YES YES YES YES YESLoan Type FE NO NO NO NO NO NO YESObservations 1,651 1,651 1,651 1,651 1,651 1,651 1,651R-Squared 0.410 0.629 0.408 0.625 0.412 0.629 0.681

Panel B. Earnings Volatility Subsamples

Bottom Tercile Bottom 35% Bottom 30% Bottom 25%Longholder CEO -0.035 -0.029 -0.049 -0.021

(-0.837) (-0.673) (-1.055) (-0.428)Longholder CFO 0.022 0.014 0.052 0.082*

(0.628) (0.390) (1.274) (1.874)Observations 550 574 492 417R-Squared 0.776 0.776 0.780 0.800

Medium Tercile Medium 35% Medium 30% Medium 25%Longholder CEO -0.109*** -0.101*** -0.106*** -0.108***

(-3.320) (-2.993) (-3.510) (-3.839)Longholder CFO -0.007 -0.001 -0.003 -0.015

(-0.227) (-0.036) (-0.101) (-0.681)Observations 547 500 662 822R-Squared 0.739 0.753 0.732 0.718

Top Tercile Top 35% Top 30% Top 25%Longholder CEO -0.024 -0.024 -0.016 -0.004

(-0.666) (-0.686) (-0.423) (-0.117)Longholder CFO 0.011 0.009 0.022 0.034

(0.397) (0.321) (0.820) (1.166)Observations 554 577 497 412R-Squared 0.777 0.770 0.787 0.831

76

Table C.6CFO Hiring

Tobit regressions with CFO overconfidence as dependent variable and CEO overconfidence asmain regressor. Both proxies are constructed following Otto (2014) and are normalized by theirsample standard deviations. Control variables and sample restrictions are as in Table 8. ***,**, and * indicate statistically different from zero at the 1%, 5%, and 10% level of significance,respectively.

(1) (2) (3) (4)Longholder CEO 0.275 0.281* 0.327* 0.343**

(1.600) (1.700) (1.917) (2.036)Manager Controls NO NO YES YESFirm Controls NO NO NO YESIndustry FE NO YES YES YESYear FE YES YES YES YESObservations 202 202 202 202Pseudo R-Squared 0.007 0.106 0.119 0.140

Table C.7Debt Issues (SDC)

Table C.7 presents the estimated log odds ratios from logit regressions with a binary variableequal to one if the firm issued debt during the fiscal year as dependent variable, conditioningon having issued debt, equity, or hybrid securities. Data on public issues are from SDC andinclude 330 firms. Equity issues are issues of common stock or non-convertible preferred stock.Debt issues are issues of non-convertible debt. Hybrid issues are issues of convertible debtor convertible preferred stock. The overconfidence proxy and the control variables are as inTable 3. Standard errors are clustered by firm, and corresponding t-statistics are shown inparentheses. ***, **, and * indicate statistically different from zero at the 1%, 5%, and 10%level of significance, respectively.

(1) (2) (3) (4) (5) (6) (7)Longholder CEO 0.716** 0.201 0.528* 0.309 -0.062

(2.537) (0.506) (1.856) (0.923) (-0.149)Longholder CFO 0.819*** 0.781** 0.688** 0.922*** 0.804**

(3.019) (2.162) (2.476) (2.801) (2.158)

Manager Controls NO YES NO YES NO YES YESFirm Controls NO YES NO YES NO NO YESIndustry FE YES YES YES YES YES YES YESYear FE NO YES NO YES NO YES YESObservations 694 611 694 587 694 598 585Pseudo R-squared 0.092 0.550 0.098 0.558 0.105 0.253 0.557

77

Table C.8Debt Issues (Compustat), Restricted Sample

Logit regressions with the Net Debt Issues Indicator as the dependent variable, regressed on our measure ofoverconfidence for CEOs and CFOs and the same control variables as in Table 3. The sample includes onlyfirms in the Restricted Sample, i.e., firms that have appreciated by more than 40% in the previous ten years(therefore excluding from the sample all the firms that, in any given year, have been listed for less than 10years). ***, **, and * indicate statistically different from zero at the 1%, 5%, and 10% level of significance,respectively.

(1) (2) (3) (4) (5) (6) (7)Longholder CEO 0.010 -0.063 -0.102 -0.103 -0.183

(0.0623) (-0.398) (-0.612) (-0.617) (-1.132)Longholder CFO 0.419*** 0.457*** 0.439*** 0.483*** 0.501***

(2.982) (3.075) (3.054) (3.098) (3.289)Manager Controls NO YES NO YES NO YES YESFirm Controls NO YES NO YES NO NO YESIndustry FE YES YES YES YES YES YES YESYear FE NO YES NO YES NO YES YESObservations 1,744 1,744 1,744 1,744 1,744 1,744 1,744Pseudo R2 0.041 0.170 0.048 0.176 0.048 0.115 0.176

Table C.9Cost of Debt Financing, Restricted Sample

Regressions of log(Interest Spread) on our overconfidence measures for CEOs and CFOs and several controlvariables (defined in Table 6), including year and industry fixed-effects. Log(Interest Spread) is the differencebetween the interest rate the borrower pays in basis points and the London Interbank Offered Rate. Thesample includes only firms in the Restricted Sample, i.e., firms that have appreciated by more than 40% inthe previous ten years (therefore excluding from the sample all the firms that, in any given year, have beenlisted for less than 10 years). ***, **, and * indicate statistically different from zero at the 1%, 5%, and 10%level of significance, respectively.

(1) (2) (3) (4) (5) (6) (7)Longholder CEO -0.237*** -0.186*** -0.266*** -0.189*** -0.135**

(-2.611) (-3.092) (-2.926) (-3.010) (-2.384)Longholder CFO 0.016 -0.050 0.089 0.012 -0.000

(0.168) (-0.851) (1.010) (0.192) (-0.004)Manager Controls NO YES NO YES NO YES YESFirm Controls NO YES NO YES NO YES YESIndustry FE YES YES YES YES YES YES YESYear-Quarter FE YES YES YES YES YES YES YESLoan Type FE NO NO NO NO NO NO YESObservations 1,162 1,162 1,162 1,162 1,162 1,162 1,162R-squared 0.479 0.668 0.468 0.664 0.481 0.674 0.711

78

Table C.10Net Interest Rates Across Subsamples (Alternative Sorting Variables)

Panels A, B and C of Table C.10 test the relation between CEO overconfidence and the cost ofdebt across different subsamples, splitting the main sample in terciles of sorting variables (Z-Score in Panel A, Leverage in Panel B, and Total Assets in Panel C). All panels show regressionsof log(Interest Rate Spread) on our measures of overconfidence and the same control variablesand fixed effects as in Column 7 of Table 6. We estimate the empirical model specified inequation 8 in the main text in each subsample. ***, ** and * indicate statistically differentfrom zero at the 1%, 5%, and 10% level of significance, respectively.

Panel A. Sorting by Z-Score(1) (2) (3)

Bottom Tercile Medium Tercile Top TercileLongholder CEO -0.153** 0.006 -0.056

(-2.016) (0.087) (-0.761)Longholder CFO -0.099 -0.060 -0.096

(-1.447) (-0.952) (-1.328)Observations 567 544 540R2 0.784 0.790 0.761

Panel B. Sorting by Leverage(1) (2) (3)

Bottom Tercile Medium Tercile Top TercileLongholder CEO -0.153* -0.060 -0.069

(-1.828) (-0.892) (-0.822)Longholder CFO -0.097 0.009 -0.095

(-1.429) (0.114) (-1.348)Observations 568 543 540R2 0.754 0.780 0.772

Panel C. Sorting by Assets(1) (2) (3)

Bottom Tercile Medium Tercile Top TercileLongholder CEO 0.063 -0.131* -0.155*

(0.900) (-1.771) (-1.759)Longholder CFO -0.148** -0.001 -0.139*

(-2.462) (-0.015) (-1.782)Observations 564 550 537R2 0.576 0.740 0.803

79

Figure C.1Coefficient Estimates with Different Transactions Thresholds

These panels report the relevant coefficient estimates from the empirical analysis of the paper (Tables 3 through8). For brevity only the coefficient from the most conservative test (last column of each table) is shown. Thex-axis reports the minimum number of transactions required for CEOs and CFOs to be included in the sample.The y-axis has the value of the estimated coefficient of interest. Panels a, b, c and d report the coefficient onLongholder CFO. Panel c reports the coefficient on Financing Deficit × Longholder CFO. Panels e, f, g, h andi report coefficients on Longholder CEO. Panels f, g, h, where we divide the sample using different measures ofvolatility, have three different coefficients. The coefficients on Longholder estimated in the low, medium andhigh volatility subsamples are plotted using dotted, solid and dashed lines, respectively.

1 2 3 4 5 6 7 8 9 100

0.10

0.20

0.30

0.40

0.50

a. Debt Issues (Compustat)

1 2 3 4 5 6 7 8 9 100

0.50

1

1.50

b. Debt Issues (SDC)

1 2 3 4 5 6 7 8 9 100

0.05

0.10

0.15

0.20

0.25

c. Financing Deficit

1 2 3 4 5 6 7 8 9 100

1

2

3

4

d. Leverage

1 2 3 4 5 6 7 8 9 10-0.15

-0.10

-0.05

0

e. Cost of Debt Financing

1 2 3 4 5 6 7 8 9 10-0.40

-0.30

-0.20

-0.10

0

f. Sorting by Earnings Vol.

1 2 3 4 5 6 7 8 9 10-0.40

-0.30

-0.20

-0.10

0

0.10

g. Sorting by An. Cov.

1 2 3 4 5 6 7 8 9 10-0.40

-0.30

-0.20

-0.10

0

0.10

h. Sorting by CV

1 2 3 4 5 6 7 8 9 100

0.50

1

1.50

2

i. CFO Hiring

80